(Evaluation Metrics)
In this literature review, a number of papers were studied between the period of 2010–2021 and a plethora of both ML and DL techniques were utilized in these papers to build or compare models to detect and classify network attacks. Table 3 presents a list of all the respected papers that utilized the different algorithms, highlighting all problem domains where each algorithm was used for as well as the highest performance achieved. Figure 1 presents the number of articles that utilized each algorithm. As seen from the figure and table, RF and SVM were the most widely used algorithms in a good number of papers and ELM was the least applied algorithm. For ML algorithms, the best performing algorithms were DT, RF, and KNN with their accuracy reaching up to 100% and the least utilized algorithms were J.48 and KNN. For DL algorithms, the best performing algorithm was RRN with the highest accuracy of 100% achieved and the least utilized and least popular algorithm was ELM, which is considered to be fast in terms of training as it consists of a single hidden layer, so it is usually applied to simple applications. However, it has recently been extended to be hierarchical to handle more complex problems with higher accuracy [ 152 ].
ML and DL algorithms used in the reviewed papers.
ML and DL algorithms evaluated in the reviewed papers.
Algorithm | Papers That Applied It | No. of Articles | Problem Domains | Performance (Highest Accuracy) |
---|---|---|---|---|
SVM | [ , , , , , , , , , , , , , , , , , , , , , , , , ] | 26 | Insider Threat, DDoS, Malware, Botnet, Malicious Traffic, IDS, Phishing | 93.95% (IDS) |
DT | [ , , , , , , , , , , , , ] | 13 | Insider Threat, DDoS, Phishing, Malware, Botnet, Malicious Traffic, IDS | 100% (Malicious Traffic) |
RF | [ , , , , , , , , , , , , , , , , , , , , , , , , , , ] | 27 | DDoS, Phishing, Malware, Botnet, IoT Network, Malicious Traffic, DNS Level Attack, IDS | 100% (Malicious Traffic, DDoS) |
NB | [ , , , , , , , , , , , , , , , , , , ] | 19 | DDoS, Malware, Botnet, Malicious Traffic, DNS Level Attack, IDS, Phishing | 90% (Malicious Traffic) |
KNN | [ , , , , , ] | 6 | Botnet, Malicious Traffic, DNS Level Attack, IDS | 100% (Malicious Traffic) |
MLP | [ , , , , , , , , , ] | 11 | DDoS, Malware, Botnet, Malicious Traffic, IDS, Phishing | 99.60% (Malware) |
ELM | [ , ] | 2 | IDS | 99.5% (IDS) |
LR | [ , , , , , , , ] | 8 | DDoS, Malware, Malicious Traffic, DNS Level Attack, IDS | 99.92% (Malware) |
J.48 | [ , , , , , ] | 6 | DDoS, IDS | 99.66% (IDS) |
ANN | [ , , , , , ] | 6 | Phishing, Zero-Day, IDS | 99.6% (Zero-Day) |
RNN | [ , , , , , , , , , ] | 10 | Insider Threat, DDoS, Malicious Traffic, IDS | 100% (Insider Threat) |
CNN | [ , , , , , , , , , ] | 10 | Insider Threat, DDoS, Malware, IoT Network, Malicious Traffic, IDS | 99% (DDoS) |
DNN | [ , , , ] | 4 | Insider Threat, DDoS, IoT Network, Malicious Traffic | 99.99% (IoT Network) |
LSTM | [ , , , , , , ] | 7 | DDoS, Botnet, IoT Network, Malicious Traffic, Phishing | 99% (DDoS) |
CNN-LSTM | [ , ] | 2 | Insider Threat, DDoS | 99.48% (DDoS) |
AE | [ , , ] | 3 | DDoS, IDS | 99% (DDoS) |
There are several datasets used by researchers in the reviewed papers to evaluate their network detection and classification model. The most widely used dataset is NSL-KDD due to the reasonable size of its training and testing sets and is also available publicly. There are 41 features in the NSL-KDD dataset. It is an enhanced version of the KDD dataset and removed the duplication of the records to eliminate the bias of the classifiers. Then, KDD-99 and CICIDS2017 came after NSL-KDD. The KDD-99 dataset was used for the first time in a competition and is an improved version of DARAP98. The CICIDS2017 dataset contains normal and new attacks and was published in 2017 by the Canadian Institute for Cybersecurity (CIC).
After that, the UNSW-NB15 dataset comes next in terms of repeatedly being used. The IXIA tool was used for creating the UNSW-NB15 dataset and it consists of nine types of attacks.
There are many other datasets, however, few researchers have tried to create their datasets. The CTU-13 dataset was captured by CTU University in the Czech Republic. It contains real botnet traffic combined with normal traffic and contains thirteen scenarios including legitimate traffic and attacks such as DoS. The SNMP-MIB dataset consists of about 4998 records with 34 variables. The attacks recorded in the data include six DoS attacks (TCP-SYN, ICMP-ECHO, HTTP flood, UDP flood, Slowloris, Slowpost) and web brute force attacks. The Kyoto 2006+ dataset was built from real traffic data from Kyoto University’s Honeypots over three years, from November 2006 to August 2009. The Kyoto 2006+ dataset consists of 24 features, 14 of which are derived from the KDD-99 dataset and 10 additional features that can be used to analyze and evaluate the IDS network. Honeypots, email server, darknet sensors, and web crawler were used to construct the Kyoto 2006+.
ADFA is an IDS that includes three data types in its structure: (1) normal training data with 4373 traces; (2) normal validation data with 833 traces; and (3) attack data with 10 attacks per vector. As the web became a significant internet criminal activity platform, the security community put in efforts to blacklist malicious URLs. Ma et al.’s dataset [ 153 ] consists of 121 sets with overall 2.3 million URLs and 3.2 million features in the dataset. The researchers divided the URLs into three groups based on their characteristics, with features being identified as binary, non-binary, numerical, or discrete.
Table 4 lists all the respected papers that utilized the different datasets, highlighting the main references for all datasets as well as the last year when each dataset was used. Figure 2 presents the number of articles that utilized each dataset.
Datasets used in the reviewed papers.
Network traffic datasets used in the reviewed papers.
Dataset | Articles | Number | Last Time Dataset Used | Publicly Available |
---|---|---|---|---|
DARPA-1998 | [ ] | 1 | 2012 | [ ] |
KDD-99 | [ , , , , , , , , , , , ] | 12 | 2018 | [ ] |
NSL-KDD | [ , , , , , , , , , , , , , , , , , , ] | 19 | 2021 | [ ] |
UNSW-NB15 | [ , , , , , , ] | 7 | 2020 | [ ] |
CICIDS-2017 or 2018 | [ , , , , , , , ] | 8 | 2020 | [ ] |
CTU-13 | [ , , , ] | 4 | 2021 | [ ] |
IoTID 20 | [ , , ] | 3 | 2021 | [ ] |
Kyoto 2006+ | [ ] | 1 | 2018 | [ ] |
CERT v6 or v4 | [ , ] | 2 | 2021 | [ , ] |
SNMP-MIB | [ ] | 1 | 2019 | [ ] |
ISCX 2012 or 2016 | [ , , , ] | 4 | 2020 | [ , ] |
ADFA | [ , ] | 2 | 2019 | [ ] |
CAIDA | [ , ] | 2 | 2016 | [ ] |
ISOT CID | [ , ] | 2 | 2021 | [ ] |
ISOT HTTP | [ , ] | 2 | 2020 | [ ] |
Malicious URLs Dataset | [ ] | 1 | 2021 | [ ] |
EMBER | [ ] | 1 | 2018 | [ ] |
CICDDoS2019 or CICDoS2017 | [ , , ] | 3 | 2020 | [ , ] |
USTCTFC2016 | [ ] | 1 | 2016 | [ ] |
GPRS WPA2/WEP | [ ] | 1 | 2017 | [ ] |
MTA KDD 19 | [ ] | 1 | 2020 | [ ] |
LITNET-2020 | [ ] | 1 | 2020 | [ ] |
CIRA-CIC-DoHBrw-2020 | [ ] | 1 | 2020 | [ ] |
Bot-IoT | [ ] | 1 | 2019 | [ ] |
Kaggle Datasets | [ , ] | 2 | 2021 | [ , ] |
UCI Datasets | [ , , ] | 3 | 2021 | [ ] |
Network security is a major concern for individuals, profit, and non-profit organizations as well as governmental organizations. In fact, with the digital explosion that we are witnessing in the present era, ensuring network security is an urgent necessity in order to safeguard society’s acceptance for thousands and thousands of services that rely essentially on the backbone of the digital life, which is the network. Therefore, network security turns out to be an urgent requirement, and not a luxury. Although many protection methods have been introduced, there are still some vulnerabilities that are exploited by hackers, leaving the network security administrators in a continuous race against the network attackers. Techniques that hover around the use of intelligent methods, namely machine learning (ML) and deep learning (DL) have proved their merits in several domains including health care systems, financial analysis, higher education, energy industry, etc. This indeed motivated the people responsible for the network security to further explore the ability of these techniques in providing the required level of network security. Consequently, several intelligent security techniques have been offered in the past few years. Although these techniques showed exceptional performance, the problem has not been resolved entirely. This leaves us in a position to critically evaluate the currently offered solutions to recognize the possible research directions that might lead to building more secured network environments.
The complication of using the right dataset and features or the right ML and DL algorithms to identify the different attack types has proven to be an arduous decision for experts to make. Hence, among the reviewed papers, some researchers focused on comparing different algorithms to determine which algorithm to use for building an intelligent model using a training dataset. As no algorithm has been found to be a silver bullet for identifying and classifying all attacks with high accuracy, it was widely noted that it is not reasonable to accept a single algorithm as a universal model.
When building any intelligent system, the designer should take into account what is/are the algorithm(s) that best fit the domain. Not only this, but the designer should also decide which dataset comprises a set of features that better represent the classification area. Considering the network attacks, this research article found that RF is the most commonly used algorithm and this can be justified due to the fact that it uses an ensemble learning technique, which to some extent might ensure a life-long system due to the exceptional capability to continuously learn new knowledge on the fly. Producing models with reduced overfitting is another motivation behind using the RF. Not only this, but RF can also be effectively applied on both categorical and continuous features, and thus it can be applied to a wide range of datasets. In addition, the exceptional ability to handle missing data puts RF as a first option when building network attack mitigation models taking into account that most of the datasets are susceptible to include missing values. However, since RF produces complex trees, building a real-life system based on RF could be a challenging task because it might require more computational power and resources, while in fact, the main success factor for building a system for detecting network attacks is the quick and instant reaction. SVM is the second most widely used algorithm. However, SVM is applied to a fewer number of network attacks when compared to RF. This can be justified due to the fact that SVM produces complex intelligent models that are difficult to apply in real life. Nevertheless, SVM is considered as the main competitor to RF due to the fact that it shares several advantages with RF such as the exceptional capability to deal with missing values, and the remarkable capability to reduce the overfitting problem. NB ranks in third place, but still did not achieve the same predictive performance as RF and SVM due to the fact that it assumes that the dataset features are independent, which in fact, is not true in most training datasets. DT was employed almost half the time that RF and SVM were used. DT proved its merits in several domains, but in the network security domains, it has not been used very much. This can be justified due to the fact that it produces a set of rules that if exposed to the attackers, they can adopt their attacks by avoiding the rules adopted from the DT models.
Included among the algorithms that conveyed excellent performing results were DL models, namely, DNN and RNN as well as ML models, namely, RF and DT with their accuracies reaching up to 100%. A more promising research direction to explore can increasingly be toward applying hybrid or ensemble models to improve attack detection accuracy; for instance, augmenting DL techniques such as CNN with long short-term memory (LSTM) for automating feature engineering and improving network attack detection accuracy. Furthermore, gated recurrent unit (GRU), initially proposed in 2014, can further be applied by researchers in solving various problem domains in network security as it is considered more efficient than LSTM, and it uses comparatively less memory, and executes faster. They can solve complex problems faster, if trained well, and therefore, they are worth trying in network attack detection, namely for DDoS or in IoT networks.
Since the performance of the intelligent models largely depend on the datasets used for training them, it is important to analyze and evaluate which dataset to use for which type of attack. It is recommended that large datasets are used with a good distribution of each class type to increase the detection and classification accuracy. Moreover, limited availability of such datasets represents a challenge in the development of more robust intelligent-based models and highlights the need for producing and publishing more new datasets in different network attack problem domains. Most of the authors in the reviewed articles used the KDD-99 dataset as well as its latest version, the NSL-KDD dataset. However, the ADFA dataset was also used by some, which was proposed as a replacement for the KDD-99 dataset, ISOT HTTP for botnet, ISOT CID for cloud environments, and IoT20 for IoT environments, so can be explored further and used to build different ML and DL models.
Identifying malicious and benign URLs was also a fundamental research direction carried out by researchers where an important set of features that affected the model accuracy were URL related features. It was found that additional improvements in classifying malicious and benign URLs can be accomplished by deploying a lexical approach, which uses static lexical features extrapolated from the URL, in addition to analyzing the URL contents for instantaneous and reliable results. Hence, using a lexical approach to classify URLs can be an important direction to explore.
Several other problem domains need to be explored as they could be a valuable direction for enhancing network security in the modern world. Namely, with the growing establishment of encrypted network traffic as well as virtual private networks, more research needs to be carried out in detecting malicious traffic in these domains using intelligent techniques as not enough research has been focused in this area. Furthermore, with the rising number of inter-connected devices and the establishments of Internet of Things (IoTs) networks, more investigation needs to be carried out in assessing different intelligent techniques on new datasets such as IoT20 as well as paving ways to developing software that can detect and analyze data packets communicated in IoT environments to update the existing datasets for more attacks. Additionally, a new protocol called DNS over HTTP (DoH) has been created recently for which more research needs to be explored on detecting malicious DoH traffic at this (DNS) level.
Finally, multiple researchers intend in their future work to convert the models they built into a real-time system in order to benefit from them in real-life scenarios such as in attack detection and prevention. There are two levels of real-time ML which are online predictions and online learning. Online prediction means making predictions in real-time. Furthermore, online learning allows for the system to incorporate new data and update the model in real-time. Hence, converting intelligent models into real time systems may be considered as a fundamental direction to probe by more researchers.
Conceptualization, M.A. (Malak Aljabri), S.S.A., R.M.A.M. and S.H.A.; methodology, M.A. (Malak Aljabri), S.S.A., R.M.A.M. and S.H.A.; software, S.M., F.M.A., M.A. (Mennah Aboulnour), D.M.A., D.H.A. and H.S.A.; validation, M.A. (Malak Aljabri), S.M. and F.M.A.; formal analysis, M.A. (Malak Aljabri), S.M. and F.M.A.; investigation, M.A. (Malak Aljabri), S.M., F.M.A., M.A. (Mennah Aboulnour), D.M.A., D.H.A. and H.S.A.; resources, M.A. (Malak Aljabri), S.M., F.M.A., M.A. (Mennah Aboulnour), D.M.A., D.H.A. and H.S.A.; data curation, S.M. and F.M.A.; writing—original draft preparation, M.A. (Malak Aljabri), S.M., F.M.A., M.A. (Mennah Aboulnour), D.M.A., D.H.A. and H.S.A.; writing—review and editing, M.A. (Malak Aljabri), S.M., F.M.A., S.S.A., R.M.A.M. and S.H.A.; visualization, S.M. and F.M.A.; supervision, M.A. (Malak Aljabri); project administration, M.A. (Malak Aljabri); funding acquisition, M.A. (Malak Aljabri) and S.S.A. All authors have read and agreed to the published version of the manuscript.
We would like to thank SAUDI ARAMCO Cybersecurity Chair for funding this project.
The authors declare no conflict of interest.
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Sei digital library, latest publications, embracing ai: unlocking scalability and transformation through generative text, imagery, and synthetic audio, august 28, 2024 • webcast, by tyler brooks , shannon gallagher , dominic a. ross.
In this webcast, Tyler Brooks, Shannon Gallagher, and Dominic Ross aim to demystify AI and illustrate its transformative power in achieving scalability, adapting to changing landscapes, and driving digital innovation.
August 27, 2024 • white paper, by nathan m. vanhoudnos , carol j. smith , matt churilla , shing-hon lau , lauren mcilvenny , greg touhill.
This paper describes counter artificial intelligence (AI) and provides recommendations on what can be done about it.
August 27, 2024 • conference paper, by rachel brower-sinning , grace lewis , sebastián echeverría , ipek ozkaya.
This paper presents an approach based on quality attribute (QA) scenarios to elicit and define system- and model-relevant test cases for ML models.
August 27, 2024 • podcast, by mckinley sconiers-hasan.
McKinley Sconiers-Hasan discusses three API risks and how to address them through the lens of zero trust.
August 20, 2024 • white paper, by allen d. householder , vijay s. sarvepalli , jeff havrilla , matt churilla , lena pons , shing-hon lau , nathan m. vanhoudnos , andrew kompanek , lauren mcilvenny.
In this paper, the authors describe lessons learned from coordinating AI and ML vulnerabilities at the SEI's CERT/CC.
July 30, 2024 • white paper, by alejandro gomez , alex vesey.
This white paper discusses the design, desired qualities, development, testing, support, and security of modern application programming interfaces (APIs).
July 26, 2024 • podcast, by jeff gennari , samuel j. perl.
Jeff Gennari and Sam Perl discuss applications for LLMs in cybersecurity, potential challenges, and recommendations for evaluating LLMs.
July 24, 2024 • podcast, by anandi hira , bill nichols.
This SEI podcast introduces capability-based planning (CBP) and its use and application in software acquisition.
July 23, 2024 • technical note, by gabriel moreno , anton hristozov , john e. robert , mark h. klein.
This report describes a model problem to support research in large-scale assurance.
June 28, 2024 • podcast, by david svoboda.
David Svoboda discusses two vulnerabilities related to Rust, their sources, and how to mitigate them.
Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Scientific Reports volume 14 , Article number: 20179 ( 2024 ) Cite this article
1 Altmetric
Metrics details
Nowadays, customer churn issues are becoming more and more important, which is one of the most important metrics for evaluating the health of a business it is difficult to measure success without measuring customer churn metrics. However, it has become a challenge for the industry to predict when customers are churning or preparing to churn and to take the necessary action at the critical time before they do. At the same time, how to keep the place of deep research on the 17 machine learning algorithms in 9 major classes of machine learning classics production is the first problem we are facing. Through customer churn deep research, we mentioned the Ensemble-Fusion model based on machine learning and introduced a smart intelligent system to help reduce the actual customer churn about the production. Comparing with most popular predictive models, such as the Support vector machine algorithm, Random Forest algorithm, K-Nearest-Neighbor algorithm, Gradient boosting algorithm, Logistic regression algorithm, Bayesian algorithm, Decision tree algorithm, and Neural network algorithm are applied to check the effect on accuracy, AUC, and F1-score. By comparing with 17 algorithms in 9 categories of machine learning classics, the data prediction accuracy of the Ensemble-Fusion model reaches 95.35%, AUC score reaches 91% and F1-Score reaches 96.96%. The experimental results show that the data prediction accuracy of the Ensemble-Fusion model outperforms that of other benchmark algorithms.
Introduction.
Customer churn is one of the key factors affecting the benign development of industries and enterprises, and at the same time, it is a very challenging research topic in both academia and industry 1 , 2 , 3 , especially for those information industries relying on the subscription model and the order purchase operation model, customer churn, especially the churn of key customers, can be fatal to their impact. Reducing 5% of customer loss rate can increase profits by 25–125% 2 . Unfortunately, this always requires lots of manual efforts to analyze data, and it is often too late to take actions to retain them. In order to retain more existing old customers, especially some key customers, many companies have made many attempts to differentiate between churned and non-churned customers, so as to achieve the purpose of retaining churned customers, but the actual effect is very poor. As we all know, the loss of old customers not only affects revenue, but also affects the attraction of new customers. In addition, the cost of developing a new customer is often much higher (almost 5–6 times) than the cost of retaining an old customer 4 , 5 . So, is it possible to research efficient customer churn prediction models for customer churn prediction by using machine learning-related algorithms in conjunction with the actual needs of the industry? At the same time, in order to help those decision makers who do not have the theoretical foundation of algorithms to make decisions quickly and efficiently, is it possible to develop an intelligent, convenient, efficient and intelligent early warning system that can detect or predict the existing customer churn in a timely manner to help the industry, and then the enterprises can take relevant actions to retain customers when they find that there is a risk of churning key customers, so as to minimize the losses of the enterprises? In part of the related work the theoretical basis of Gradient Boosting Algorithm 6 , 7 , Bayesian Algorithm 8 , 9 , Support Vector Machine Algorithm 10 , 11 , 12 , 13 , 14 , 15 , Random Forest Algorithm 16 , K Neighborhood Algorithm 17 , 18 , Logistic Regression Algorith 19 , 20 , Decision Tree Algorithm 21 , 22 , 23 , 24 and Neural Network Algorithms 25 , 26 , 27 , 28 , 29 are described and the research on application of these algorithms in customer churn prediction is discussed. The literature related to the above algorithms is restating the superiority of the single algorithm they use, and after analyzing them, it can be concluded that these algorithms are affected by the characteristics of the dataset, and there is a strong dependency between their algorithms and the dataset, and then there is no such thing as being able to use one algorithm alone to solve all the problems in any practical application scenarios. Based on the shortcomings of the traditional algorithms analyzed above, this paper proposes a model based on Ensemble-Fusion (Integrated Learning Fusion), in order to meet the universality of various complex scenarios through the model, and expects to be able to provide academia and industry with a pervasive and efficient customer churn prediction solution. So in this paper, we first propose a customer churn prediction algorithm based on the Ensemble-Fusion model. Then it proposes an efficient churn solution based on the Ensemble-Fusion model. Finally, in order to help the information industry make efficient customer churn decisions, a real-time intelligent early warning system for customer churn is developed through theory-guided practice, which can monitor customer dynamics in real-time, help enterprises to identify potential lost customers in advance, and provide early warning at the first moment to remind the sales team or the Customer success management team (CSM) to take proactive action to retain lost customers, thus reducing the risk of fatal blow to the enterprise because of customer churn.
Given the above purposes, this paper conducts research on customer churn prediction through machine learning related theories and algorithms, firstly gives a solution to deal with the huge and complex datasets in the industry, then proposes the Ensemble-Fusion (Integrated Learning Fusion) prediction model for customer churn, and finally, in order to further guide the theory to practice, facilitate the enterprises to take actions quickly and efficiently to retain customers, especially the key customers, in order to improve customer retention. Especially the retention of key customers. Combined with my many years of experience in the industry, I have developed an end-to-end real-time intelligent early warning system for customer churn, which not only predicts customer churn in an organization’s production environment, but also sends out early warnings to alert the relevant personnel such as the sales team and the customer success team, so that the relevant teams can take effective action to retain the customers who are about to be lost in the first time. The system not only predicts customer churn in an organization’s production environment, but also sends out early warnings to alert relevant personnel such as sales and customer success teams so that they can take immediate action to retain lost customers. In order to solve the above problems, we must first deal with the problems encountered in the research, specifically in the research work encountered in the actual research and development of the very difficult problems are as follows: First, the real structure of the production data is very complex and the relevant data are often distributed in different regions of the world in different departments and data structure of different databases, the collection of data is very difficult, and due to the restriction of some sensitive information and the relevant agreements, it is very difficult to collect all the relevant data. It is also difficult to collect all the relevant data due to sensitive information and related protocol issues. Therefore, the problem of customer churn data collection becomes how to construct an effective model with a limited data set. Secondly, in the collected relevant data, there is still a lot of noise in the data, which is very imbalance 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 due to the actual impact of business complexity and there are no labels to mark whether a customer is churned or not, which requires that a lot of prior work and business knowledge should be involved before proceeding with the collection and processing of the data. In order to address the above issues in customer churn data prediction, this paper’s main contributions of the work are as follows:
This paper proposes a novel model named Ensemble-Fusion based on ML (Machine Learning) related theories and algorithms to predict customer churn in SAAS 36 (Software-as-a-Service, SAAS is a cloud-based software delivery model in which the cloud provider develops and maintains cloud application software) production environments, which focuses on the exceptionally complex data collection, processing and application in the actual production line, and organizes a detailed customer churn prediction data processing architecture diagram is shown(detailed in Sect. “ Customer churn prediction solution based on Ensemble-Fusion model ”), and finally the solution proposed in this paper is used in the actual production environment to achieve good results.
This paper combines machine learning theories and algorithms, such as support vector machine algorithms, random forest algorithms, K-neighborhood algorithms, gradient boosting algorithms, logistic regression algorithms, Bayesian algorithms, deci- sion tree algorithms and neural network algorithms, and other 9 categories of 17 machine learning algorithms as a baseline classifiers to propose the “customer churn data processing architecture based on the integration of learning fusion (Ensemble- Fusion)”. Fusion-based customer churn prediction model and verified the high accuracy and effectiveness of the churn prediction model by evaluating the key indexes of the machine learning model, such as precision, recall, accuracy, AUC 37 (Area under the ROC 38 Curve, AUC measures the entire two-dimensional area underneath the entire ROC curve. AUC provides an aggregate measure of performance across all possible classification thresholds.) and F1-score 39 , 40 (F1-score is an important evaluation metric that is commonly used in classification task to evaluate the performance of a model. F1-score is a way of combining the precision and recall of the model, and it is defined as the harmonic mean of the model’s precision and recall).
In order to further improve the productivity of the industry efficiently, by linking theory to practice, this paper also designs and develops an intelligent early warning system based on the Ensemble-Fusion model to help enterprises predict customer churn, especially the churn of important customers, quickly and effectively, so as to help them retain churned customers and reduce the churn that brings. The system is designed to help companies retain lost customers and minimize the fatal blow to the company due to customer churn. The intelligent system can not only present important customers with high probability of churn, but also automatically provide relevant information based on the prediction results to remind relevant personnel to take proactive actions to retain important customers that are about to be churned, so as to reduce losses.
This paper not only provides specific solutions to the important problem of cus- tomer churn from theory, but also translates the theory into a specific intelligent early warning system, which can efficiently help enterprises, especially those who don’t know the background knowledge of machine learning and other relevant leadership decision- making personnel to easily make effective decisions about customer churn, so as to be able to retain key customers and increase the competitiveness of the enterprise. The system can be used to retain key customers and increase the competitiveness of an organization.
The rest of this paper is organized as follows, in Section “ A research approach to customer churn prediction based on Ensemble-Fusion model ”, it mainly introduces the theory and methodology, solution, and overall architectural design of the machine learning-based customer churn intelligent system and introduces the customer churn prediction algorithm based on the Ensemble-Fusion model proposed in this paper. In Section “ Experiment and result ”, the proposed customer churn prediction algorithm is validated and the high accuracy and effectiveness of the churn prediction model are verified by the key metrics of machine learning model evaluation, such as precision, recall, accuracy, AUC , and F1-score 37 , 38 , 39 , 40 . Section “ Intelligent early warning system for customer churn prediction based on Ensemble-Fusion model ” describes the main functions of the intelligent early warning system for customer churn prediction, and also provides a detailed description of the User Cases associated with this intelligent system. A review of relevant customer churn research is presented in Section “ Related work ”. Finally, relevant conclusions and outlook are summarized in Section “ Conclusions and future work ”.
This part proposes a solution for customer churn prediction based on the Ensemble- Fusion model: firstly, it comprehensively outlines the specific scenarios to be solved for customer churn, and gives the ideas and feasible solutions to solve the problem from top to bottom. Then the specific design and implementation of an end-to-end customer churn intelligent prediction system is proposed: specifically including the collection and processing of complex datasets, the construction of prediction models, and the intelligent system platform in three parts, each of which contains a detailed process. Then this paper provides an in-depth analysis of the machine learning model for customer churn prediction, and finally this paper proposes a new customer churn prediction model and gives a specific implementation algorithm.
This part proposes a solution based on the Ensemble-Fusion model to predict customer churn and help organizations reduce customer churn. The detailed process of the solution is depicted in Fig. 1 , as shown in Fig. 1 , the solution consists of two main parts: the offline training part and the online inference part. During offline training, data preprocessing 30 , 31 , 32 , 33 s first required to clean and label the input data, the annotation is done by labeling the data with churn or non-churn. Then, the relevant features of the data are extracted based on the business knowledge, such as the feature “Trend of meetings compared to last year” which is used to describe the number of meetings booked by customers in the current year compared to the number of meetings booked by customers in the previous year, and the number of meetings booked also reflects the trend of imminent churn of customers. The feature “Trend in meeting duration compared to last year” can be used to characterize the total duration of meetings in the current year compared to the total duration of meetings in the previous year, which can be used to predict the trend of customer churn. These extracted features can effectively reflect the trend of imminent or significant customer churn. Specific model features are described in Table 1 , where model training data information is used from actual production line usage data.
Customer Churn Solution Flowchart.
The process of customer churn prediction processing and the logical relationship between data transfers are detailed in Fig. 2 . In addition, since there are only a few churned (noisy) data, data balancing-related processes must be performed before training. These features can then be used to iteratively train and validate the machine learning model until the model is validated well enough to be deployed directly to a production environment. Finally, the rigorously validated model can be deployed in a production environment to predict the likelihood of customer churn in real time.
Architecture diagram of customer churn prediction data processing.
For the online inference component, data cleaning and feature engineering 35 , 36 , 37 are also required to construct the training dataset. The dataset here does not contain labeled data, mainly because the goal to be predicted is whether customers will churn in the following months, which has not occurred in the previous inference process. After obtaining the trained model, test data also needs to be fed into the machine learning model to infer the final prediction. Finally, information about the high churn customers predicted by the validated machine learning model will be displayed on the intelligent churn prediction system. Information about the churn prediction will be notified to the project stakeholders in real-time via email, instant messaging, and other messaging channels so that they can proactively take action to minimize the risk of churn losses.
In order to better carry out the research on customer churn rate, this paper focuses on the theoretical basis of the Support Vector Machine algorithm, Random Forest algorithm, K-neighborhood algorithm, Gradient Boosting algorithm, Logistic Regression algorithm, Bayesian algorithm, Decision Tree algorithm, and Neural Networks algorithm in Section “ Related work ” and discusses the research on the application of these algorithms in the prediction of customer churn rate. The literature related to the above algorithms restates the superiority of the single algorithm they use, and after analyzing them, it can be concluded that these algorithms are affected by the characteristics of the dataset, and there is a strong dependency between their algorithms and the dataset, and then there is no such thing as being able to use one algorithm alone to solve all the problems in any practical application scenarios. Based on the shortcomings of the traditional algorithms analyzed above, this paper proposes a model based on Ensemble-Fusion (Integrated Learning Fusion), in order to meet the universality of various complex scenarios through the model, and expects to be able to provide academia and industry with a pervasive and efficient customer churn prediction solution.
This subsection focuses on the detailed construction process of the customer churn prediction method based on the Ensemble-Fusion model, which is described in detail in Algorithm 1, and compared with the experimental results of 17 machine learning algorithms through the model in the experimental part of Section “ Experiment and result ”, so as to validate that the model has a high accuracy rate, strong robustness, and ease of scalability.
To further help organizations reduce customer churn, this subsection designs and develops a customer churn intelligent prediction system. The system consists of three main parts, the first part is mainly the collection and processing of different business-related data set and detailed processing, which mainly includes four major processes, of which the first major process includes the access of heterogeneous data, due to the unusual complexity of the source of data in the real production environment, which mainly includes the system application data, Billing (financial billing) customer data, prod- uct transaction data, Product discount data, product sales data, cross-departmental transaction data, reconciliation data and posting data. In a large multinational group.
Customer ChurnPrediction Algorithm Based on Ensemble Fusion Model
of companies, due to the different technical architectures of each system, the data for- mat is not the same, generally JSON, XML, plain text files and other formats. To process the data, it is necessary to unify the data format here, from different hetero- generous databases through ETL (Extra, Transform, Load) to achieve from different types of databases (e.g., MySQL, Oracle, MongoDB, and Redis) to get the data, and finally unified storage in the MySQL database. The second major process is to structure the data by managing the database to construct training and testing datasets for the next machine learning models. The third major process is to perform the construction of the machine learning model for customer churn prediction through the formatted and unified dataset acquired in the previous step (details will be elaborated in Sect. “ AUC results and analysis ”). The fourth major part is the transfer of business logic through the standardized API interface (Restful API), and ultimately display of relevant information on the front-end page, which mainly includes the display of customer churn information, the display of customer churn heat map, the customer churn management platform, and the analysis of customer churn 360-degree related information, which is elaborated in detail in Fig. 2 (Customer Churn Prediction Data Processing Architecture Diagram). The second part is the ML (Machine Learning) modeling system, which includes data acquisition, feature engineering, and model training, and this part is elaborated in subsection 2.3. The third part is the visualization and presentation plat- form which will display the information related to customer churn, and this relevant part will be described in detail in Section “ Experiment and result ”. The details of the system architecture are described in detail in Fig. 3 , as shown in Fig. 3 , the system mainly consists of the following parts, the first part is the collection of data, for the Fortune 500 multi- national corporations, their various businesses are spread all over the world, and the collection of data is a very complex and time-consuming work. The second part is the data processing such as feature engineering on the data collected in the first part, then the training and validation of the machine learning model, and finally obtaining a machine learning model with the highest accuracy rate to be used in the customer churn prediction system. The third part is the platform display part, which mainly displays multi-dimensional warning information and real-time forecasts for specific customer churn information, and the specific related information and functions will be elaborated in Section “ Experiment and result ”.
Architecture diagram of customer churn intelligent early warning system.
Specific user usage examples of this intelligent system are described in detail in Fig. 4 . As shown in Fig. 4 , the sales layer and the leadership layer are two important key target roles that are important in the platform. At the sales level, the intelligent system displays customers with high churn risk on the platform and provides relevant details. The platform also sends out regular alert emails, timely messages, and other early warning information to notify the relevant project stakeholders to take proactive action to intervene in the impending churn. Additionally, salespeople can send feedback about forecasts to help continuously improve and optimize the proposed machine learning model. For leadership, it is even more important to keep track of global customer churn rather than individual customer churn. To solve this problem, the intelligent real-time alert system is designed with a dashboard module for leadership managers to show the overall churn trend from a global perspective, thus facilitating decision-makers to make efficient decisions at the first time.
Use case diagram for a customer churn platform.
This section focuses on the comparison of the experimental results of the proposed Ensemble-Fusion model-based machine learning for customer churn prediction and the classical machine learning 9 categories and 17 algorithms for customer churn predic- tion. Here, a private dataset of the customer production line system of the Company from 2015 to 2022 is used, where 80% of the data is used for training and 20% of the data is used for testing, in which K-fold cross-validation is used to test the accuracy of the model.
In order to evaluate the performance of machine learning models, relevant metrics recognized in the field of machine learning are usually used, namely precision, recall, accuracy and F1-score 38 , 39 , 40 , 41 . These metrics represent the performance of predictive models for customer churn prediction. The meanings of the metrics are explained here in a relevant way, with true positives and false positives denoted as TP and FP, respectively 42 , and true negatives and false negatives denoted as TN and FN, respectively 43 .TP stands for the number of customers whose actual labels are churned ( predict label is churn), FP stands for the customers whose actual customers are labeled as not churned but whose predicted customer labels are churned number, FN represents the number of customers whose actual label is churn but whose predicted label is not churn, and TN represents the number of customers whose actual label is not churn and whose predicted label is not churn. Thus, precision, recall, accuracy, and F1 score can be described as follows:
To evaluate the performance of the customer churn prediction algorithm based on the Ensemble-Fusion model proposed in this paper, the customer churn prediction is performed by the model proposed in this paper and 17 machine learning algorithms in 9 major categories of machine learning classics respectively. The performance metrics of precision, recall, accuracy, and F1-score 38 , 39 , 40 , 41 are compared, and the detailed results of the specific comparison can be found in Table 2 . Among the 17 machine learning algorithms in 9 major classes of machine learning classics, the accuracy of gradient boosting classifiers and random forests are 95.32% and 94.29%, respectively, and the F1-score of the gradient boosting classifier is up to 96.3%, which is better than other machine learning classic algorithmic classifiers, while the integrated learning fusion model proposed in this paper achieves an accuracy rate of 95.35%, and the F1-Score reaches 96.96% significantly better than other machine learning classic benchmark classifier algorithms. The results of Precision, Recall, Accuracy, and F1-Score of 17 machine learning algorithms in 9 categories of machine learning classics are shown in detail in Figs. 5 , 6 , 7 and 8 for comparison.
Comparison of algorithm precision.
Comparison of algorithm recall.
Algorithm Accuracy comparison chart.
Algorithm F1-Score Comparison chart.
To further evaluate the performance of the model, this section also uses AUC 13 curve for evaluating the machine learning model. A higher AUC score represents better performance of the model. Here, fivefold cross-validation 14 is used to calculate the ROC, and the highest AUC is obtained for the integrated learning-based fusion model proposed in this paper, the detailed results of the specific comparison can be found in Table 3 , and the ROC 15 results for the related machine algorithms are shown in Figs. 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 and 27 .
SVM(RBF)algorithm ROC and AUC.
SVM(Poly) algorithm AUC.
SVM (Sigmoid) algorithm AUC.
Random Forest algorithm AUC.
KNN algorithm AUC.
LR algorithm AUC.
MLP (Algorithm 16) AUC.
MLP (Algorithm 17) AUC.
MultinomialNB algorithm AUC.
BernouiliNB algorithm AUC.
GaussianNB algorithm AUC.
DT(CART) algorithm AUC.
ID3 algorithm AUC.
ExtraTrees algorithm AUC.
AdaBoost algorithm AUC.
Comparison of K-fold AUC for each algorithm.
Comparison of average AUC by algorithm.
In this section, the main functions of the real-time intelligent early warning system for customer churn data prediction based on the Ensemble-Fusion model will be elaborated in detail, and the relevant descriptions of the main functions are described as follows.
Figure 28 shows the top five of the “Top 100” accounts with high churn risk, as shown in Fig. 28 , with detailed information (e.g., account name, account ID, etc.) displayed in the table. If the prediction is incorrect, the user can give feedback by clicking on the relevant action, and then feedback through the system. Of course, it is also possible to click on the Account ID to enter the detailed prediction page, which will be analyzed in detail in Section “ Demonstration of the intelligent system of customer churn prediction ”.
Example display of customer churn information.
In Figs. 29 and 30 , detailed information of a detailed page of a real-time intelligent prediction system for customer churn is described, which consists of two parts, wherein the upper half of the page displays the basic information of the current churned customer data prediction, which specifically includes information such as the user’s ID, name, and the type of platform. In the second half, the reasons for the churn are provided and a multi-dimensional analysis of the specific reasons is provided to help the relevant stakeholders and personnel in the relevant departments in the industry to analyze the current billing and usage trends of the account so as to identify the churn trends in time to take effective action.
Example display of lost customer details.
Example display of user and account trends.
For dashboards designed for leadership decision makers, specific information about the results of predictive analysis of relevant customer churn data is presented in Figs. 31 , 32 , 33 and 34 . The Real-Time Intelligent Alerts dashboard consists of a total of five sections. The first section is the overall trend in customer churn, which includes three parts: average churn rate, fully renewed accounts, and new onboarding contracts. The second section is Customer churn as a key driver for leading decision-making teams to make decisions. The third section is the Churn heatmap (Churn Heatmap Description), which displays churn rates for selected regions and also provides a top correlation analysis and top correlation forecast for the next six months.
Leadership Decision Panel Design—Generalized Information.
Leadership Decision Panel Design—Churn Heat Map 44 (We developed a customer churn intelligent early warning system using open source pyecharts, https://github.com/pyecharts/pyecharts ).
Leadership decision panel design—correlation coefficient analysis.
Leadership decision panel design—360 degree information analysis presentation.
In order to evaluate the performance of the model in the intelligent early warning system for customer churn based on the Ensemble-Fusion model, this subsection tests the 2018 production line production data. Figure 35 demonstrates the specific results of the evaluation, and the accuracy of the model is obtained by testing and validation to be above 95.8%, which achieves a high level of accuracy prediction. Higher accuracy means that more predicted churned customers are indeed likely to actually churn in the future, which does reduce the churn rate and retention of customers thus reducing the risk of fatalities to the organization due to customer churn.
Customer churn prediction model evaluation page.
To obtain the best model for customer churn prediction, this section will conduct a theoretical analysis of related machine learning algorithms and models. First, 9 categories and 17 algorithms related to machine-learning are expounded, and then in the third part, a prediction model of customer churn rate based on an ensemble-fusion model is proposed, and 17 sets of experiments are carried out to verify that the model has strong performance. Robust and easy to extend.
Support vector machines(SVM) 10 , 11 are a set of supervised learning methods used for classification, regression, and outlier detection 12 . The advantages of support vector machines are effective in high dimensional spaces. Still effective in cases where the number of dimensions is greater than the number of samples. The objective function:
SVM is a supervised learning models that analyze data used for classification and regression analysis. In the customer churn prediction, SVM divides the result of prediction into two parts, such as positive is customer churn while negative is customer non-churn. The kernel of SVM is used like linear, poly and RBF.
Random forests are constructed by several trees 16 and each decision tree is trained by random samples. A random forest is a data construct applied to machine learning that develops large numbers of random decision trees analyzing sets of variables. This type of algorithm helps to enhance the ways that technologies analyze complex data. The Random Forest algorithm is one of the best algorithms for classification. RF can classify large data with accuracy. It is a learning method in which the number of decision trees is constructed at the time of training and outputs of the modal predicted by the individual trees. RF acts as a tree predictor where every tree depends on the ran- dom vector values. The basic concept behind this is that a group of “weak learners” may come together to build a “strong learner”. Random forest models are machine learning models that make output predictions by combining outcomes from a sequence of regression decision trees. Each tree is constructed independently and depends on a random vector sampled from the input data, with all the trees in the forest having the same distribution. The predictions from the forests are averaged using bootstrap aggregation and random feature selection. RF models have been demonstrated to be robust predictors for both small sample sizes and high dimensional data. RF clas- sification models were constructed that directly classified bioreactor runs as having sufficient or insufficient cardiomyopathy content.
K-nearest-neighbors algorithm (KNN) is a non-parametric classification method first developed by Evelyn Fix and Joseph Hodges in 1951 17 . It is used for classification and regression. In both cases, the input consists of the k closest training examples in the data set. The output depends on whether KNN 18 is used for classification or regression. The training examples are vectors in a multidimensional feature space, each with a class label. The training phase of the algorithm consists only of storing the feature vectors and class labels of the training samples. The k-nearest neighbors (KNN) algorithm is a simple, easy-to-implement supervised machine learning algorithm that can be used to solve both classification and regression problems. The principle behind nearest neighbor methods is to find a predefined number of training samples closest in distance to the new point and predict the label from these. The number of samples can be a user-defined constant (k-nearest neighbor learning) or vary based on the local density of points (radius-based neighbor learning). The distance can, in general, be any metric measure: standard Euclidean distance is the most common choice. Neighbors- based methods are known as non-generalizing machine learning methods since they simply “remember” all of their training data. KNN is a non-parametric algorithm, which means it does not make any assumptions on underlying data. It is also called a lazy learner algorithm because it does not learn from the training set immediately instead it stores the data set and at the time of classification, it performs an action on the data set. KNN algorithm at the training phase just stores the data set and when it gets new data, then it classifies that data into a category that is much similar to the new data.
Gradient boosting[34, 35]produces a model in the form of an ensemble of the prediction model, usually there using decision trees. Gradient boosting classifier has a lot of advantages, such as high prediction rate, dealing with non-linear data, and flex- ible handling of various types of data. Predictions are made by the majority vote of the weak learners’ predictions, weighted by their individual accuracy. Gradient boosting machines are an extremely popular machine learning algorithm that has proven successful across many domains. A simple GBM model contains two categories of hyper-parameters: boosting hyper-parameters and tree-specific hyper-parameters. Gradient boosting re-defines boosting as a numerical optimization problem where the objective is to minimize the loss function of the model by adding weak learners using gradient descent. Gradient descent is a first-order iterative optimization algorithm for finding a local minimum of a differentiable function. As gradient boosting is based on minimizing a loss function, different types of loss functions can be used resulting in a flexible technique that can be applied to regression, multi-class classification.
Logistic regression[19, 20]is a generalized linear regression analysis model, which is divided from the classification of machine learning. It belongs to the classification algorithm in supervised learning. Due to the good performance of logistic regression 19 , it can often be used for binary classification. or multi-classification problems. In the research on the prediction of customer churn rate, logistic regression can be abstracted here to deal with the binary classification problem, such as the label of marking customer churn as 0, and the label of non-churn as 1. At this time, for each set of input data, according to the Sigmoid function 20
in logistic regression, the predicted value can be mapped to between [0, 1]. If y ≥ 0 . 5, it is recorded as 0 category is the loss, and similarly, it is 1 category that is not lost.
The research on customer churn prediction is currently limited to the application stage of Naive Bayes 8 . The basic idea of the Naive Bayes algorithm 9 : for a given category to be classified, solve the problem under the condition that this category appears. The probability of occurrence of each category, which category has the highest probability of occurrence, is considered to be the category to which the item to be classified belongs.
In the research on customer churn prediction, a few pieces of literature use a decision tree algorithm 21 . A decision tree is also called a decision tree in some literature 22 . This kind of algorithm belongs to supervised learning in machine learning, which can be used to solve classification and regression problems. The decision tree algorithm is a top-down divide-and-conquer strategy, a recursive algorithm from the root node to the leaf node, where the leaf nodes are divided according to different division methods, generally according to information gain, gain rate, and Gini index 23 .The decision tree is divided the algorithms are ID3 algorithm, C4.5 algorithm and CART algorithm 24 .
In recent years, deep learning has been widely used to solve some complex problems, and it is also used in the prediction of customer churn rate 25 . The BP neural network was proposed by a group of scientists led by Rumelhart and McCelland in the book “Parallel Distributed Processing” in 1986, which detailed the error back-propagation algorithm for multilayer perceptions with nonlinear continuous transformation functions. The analysis of, realizes Minsky’s vision of multi-layer network 26 . The structure of BP neural network 26 is a backpropagation (Back Propagation) neural network, referred to as the BP neural network. The standard BP neural network is divided into three layers, namely the input layer, the hidden layer and the output layer, as shown in Fig. 36 .
The structure of three-layer BP neural network.
The principle of the neural network algorithm mainly includes two stages: (1)FP (forward propagation) data is input from the input layer, then input through the hidden layer under the mapping of the relevant activation function, and finally reaches the output layer for output, and then according to the error between the expected output and the actual output is used to construct the cost function (loss function) for the second stage (2) BP (backpropagation) from the output layer through each hidden layer to correct the weight and bias of the hidden layer by layer, and finally correct the weights and biases from the hidden layer to the input layer, and finally get the neural network model. Neural networks can approximate any nonlinear function arbitrarily. Because of their simple structure and easy implementation, they have been widely used in time series analysis and nonlinear function regression estimation. However, the development of such networks is limited due to the difficulty of determining the network structure, the existence of over-learning, and the tendency to fall into local extreme values. This paper expects to use it in the research of customer churn prediction to get good results.
In this paper, we proposed a novel model named Ensemble-Fusion that utilized 9 categories of 17 machine learning algorithms as baseline classifiers. Through experiment proves that the Ensemble-Fusion model (Our model) reaches 95.35%, AUC score reaches 91% and F1-Score reaches 96.96%, and the experimental results show that the data prediction accuracy of Ensemble-Fusion model outperforms that of other benchmark algorithms. This paper first elaborates on the important role of research in today’s information industry and gives important contributions, then this paper focuses on the research of customer churn prediction based on an integrated learning fusion model, mainly from the customer churn prediction solution based on the integrated learning fusion model, the design of real-time intelligent early warning system of customer churn, the machine learning algorithm of customer churn prediction and this paper. The newly proposed customer churn prediction model is compared and the specific implementation algorithm based on the integrated learning fusion model is given. Then this paper validates the proposed churn prediction algorithm experimentally and evaluates the robustness of the algorithm by using evaluation metrics such as precision, recall, accuracy, F1-score, and AUC. Finally, this paper provides a detailed description of the main functions of the theoretically and practically developed customer churn intelligent early warning system, in order to efficiently help the information industry improve its productivity and to be able to excel in today’s globally competitive environment.The study presented in this paper is not free of limitations. Firstly, it is challenging to gather all relevant data on customer churn due to sensitive information and related protocol issues. Therefore, how to construct an effective model using the limited dataset becomes a bottleneck in customer churn prediction research. The other limitation of the study is that there is still a lot of noise and no labels to mark customer churn in the collected data, which requires a lot of time to organize and learn relevant business knowledge before data collection and processing. Finally, customer churn is a multidisciplinary issue involving a variety of fields such as psychology, sociology, and economics, but current research may lack an interdisciplinary perspective and approach.Concerning future research, we intend to develop a similar ensemble-fusion classification algorithm that substitutes the baseline classifiers with reinforcement learning model-related algorithms. The primary aim here is to construct an ensemble classifier that can more easily be used in complex data structures such as multisource isomerization. In order to study customer churn in more depth in the future, there are several potential directions for further research. The first direction is to obtain more data from industry, e.g., combining different feature data. Another interesting direction is to relax strict algorithmic constraints to support compact and dense feature representations, which can be explored in areas such as fast symmetric decomposition techniques.
Availability of data and materials: Data is available on request from the author (Chenggang He).
Fujo, S. W. et al. Customer churn prediction in telecommunication industry using deep learning. Inf. Sci. Lett. 11 (1), 24 (2022).
Google Scholar
Xie, Y., Li, X., Ngai, E. & Ying, W. Customer churn prediction using improved balanced random forests. Expert Syst. Appl. 36 (3), 5445–5449 (2009).
Article Google Scholar
De Caigny, A., Coussement, K. & De Bock, K. W. A new hybrid classification algorithm for customer churn prediction based on logistic regression and decision trees. Eur. J. Oper. Res. 269 (2), 760–772 (2018).
Article MathSciNet Google Scholar
Ahmad, A. K., Jafar, A. & Aljoumaa, K. Customer churn prediction in telecom using machine learning in big data platform. J. Big Data 6 (1), 1–24 (2019).
He, C., Ding, C.H., Chen, S., Luo, B. Intelligent machine learning system for predicting customer churn. In: 2021 IEEE 33rd International Conference on Tools with Artificial Intelligence (ICTAI) , pp. 522–527 (2021). IEEE.
Estran, R., Souchaud, A. & Abitbol, D. Using a genetic algorithm to optimize an expert credit rating model. Expert Syst. Appl. 203 , 117506 (2022).
Deng, Z., Huang, Z.-H. & Miao, X. Sufficient conditions for judging quasi-strictly diagonally dominant tensors. Comput. Appl. Math. 42 (1), 63 (2023).
Chen, Y., Matsubara, T., Yaguchi, T. Kam theory meets statistical learning theory: Hamiltonian neural networks with non-zero training loss. In: Proceedings of the AAAI Conference on Artificial Intelligence , vol. 36, pp 6322–6332 (2022).
Bhavan, A. et al. Bagged support vector machines for emotion recognition from speech. Knowl.-Based Syst. 184 , 104886 (2019).
Sadohara, R. et al. Seed coat color genetics and genotype× environment effects in yellow beans via machine- learning and genome-wide association. Plant Genom 15 (1), 20173 (2022).
Varshney, R. K. et al. Resequencing of 429 chickpea accessions from 45 countries provides insights into genome diversity, domestication and agronomic traits. Nat. Genet. 51 (5), 857–864 (2019).
Article CAS PubMed Google Scholar
Bergström, A. et al. Insights into human genetic variation and population history from 929 diverse genomes. Science 367 (6484), 5012 (2020).
Keramati, A., Ghaneei, H. & Mirmohammadi, S. M. Developing a prediction model for customer churn from electronic banking services using data mining. Financ. Innov. 2 , 1–13 (2016).
Hudaib, A. et al. Hybrid data mining models for predicting customer churn. Int. J. Commun. Netw. Syst. Sci. 8 (05), 91 (2015).
Li, H., Yang, D., Yang, L., Lu, Y., Lin, X. Supervised massive data analysis for telecommunication customer churn prediction. In: 2016 IEEE International Conferences on Big Data and Cloud Computing (BDCloud), Social Comput- ing and Networking (SocialCom), Sustainable Computing and Communications (SustainCom)(BDCloud-SocialCom-SustainCom), pp 163–169 (2016). IEEE.
Deng, Q. & Söffker, D. A review of hmm-based approaches of driving behaviors recognition and prediction. IEEE Trans. Intell. Vehic. 7 (1), 21–31 (2021).
Graham, S. E. et al. The power of genetic diversity in genome-wide association studies of lipids. Nature 600 (7890), 675–679 (2021).
Article ADS CAS PubMed PubMed Central Google Scholar
Shen, J. et al. Identification of a novel gene signature for the prediction of recurrence in hcc patients by machine learning of genome-wide databases. Sci. Rep. 10 (1), 4435 (2020).
Devriendt, F., Berrevoets, J. & Verbeke, W. Why you should stop predicting customer churn and start using uplift models. Inf. Sci. 548 , 497–515 (2021).
Wang, Q.-F., Xu, M. & Hussain, A. Large-scale ensemble model for customer churn prediction in search ads. Cognit. Comput. 11 , 262–270 (2019).
Alboukaey, N., Joukhadar, A. & Ghneim, N. Dynamic behavior based churn prediction in mobile telecom. Expert Syst. Appl. 162 , 113779 (2020).
Wang, S., Cao, J. & Philip, S. Y. Deep learning for spatio-temporal data mining: A survey. IEEE Trans. Knowl. Data Eng. 34 (8), 3681–3700 (2020).
Zdravevski, E., Lameski, P., Apanowicz, C. & Ślȩzak, D. From big data to business analytics: The case study of churn prediction. Appl. Soft Comput. 90 , 106164 (2020).
Vo, N. N., Liu, S., Li, X. & Xu, G. Leveraging unstructured call log data for customer churn prediction. Knowl.-Based Syst. 212 , 106586 (2021).
Goecks, J., Jalili, V., Heiser, L. M. & Gray, J. W. How machine learning will transform biomedicine. Cell 181 (1), 92–101 (2020).
Article CAS PubMed PubMed Central Google Scholar
Aria, M., Cuccurullo, C. & Gnasso, A. A comparison among interpretative proposals for random forests. Mach. Learn. Appl. 6 , 100094 (2021).
Alotaibi, M. Z. & Haq, M. A. Customer churn prediction for telecommunication companies using machine learning and ensemble methods. Eng. Technol. Appl. Sci. Res. 14 , 14572–14578 (2024).
Alabdulwahab, A., Haq, M. A. & Alshehri, M. Cyberbullying detection using machine learning and deep learning. Int. J. Adv. Comput. Sci. Appl. 14 , 10 (2023).
Haq, M. A., Khan, M. A. & Alshehri, M. Insider threat detection based on NLP word embedding and machine learning. Intell. Autom. Soft Comput. 33 , 619–635 (2022).
Oksuz, K., Cam, B. C., Kalkan, S. & Akbas, E. Imbalance problems in object detec- tion: A review. IEEE Trans. Pattern Anal. Mach. Intell. 43 (10), 3388–3415 (2020).
Zidan, M. A. et al. A general memristor-based partial differential equation solver. Nat. Electron. 1 (7), 411–420 (2018).
Devriendt, F., Berrevoets, J. & Verbeke, W. Why you should stop predicting cus- tomer churn and start using uplift models. Inf. Sci. 548 , 497–515 (2021).
Shirazi, F. & Mohammadi, M. A big data analytics model for customer churn prediction in the retiree segment. Int. J. Inf. Manag. 48 , 238–253 (2019).
Amin, A. et al. Cross-company customer churn prediction in telecommunication: A comparison of data transformation methods. Int. J. Inf. Manag. 46 , 304–319 (2019).
Stripling, E., Broucke, S., Antonio, K., Baesens, B. & Snoeck, M. Profit maximizing logistic model for customer churn prediction using genetic algorithms. Swarm Evolut. Comput. 40 , 116–130 (2018).
Liu, Z. et al. Extreme gradient boosting trees with efficient Bayesian optimization for profit-driven customer churn prediction. Technol. Forecast. Soc. Change 198 , 122945 (2024).
Chicco, D. & Jurman, G. The advantages of the matthews correlation coeffi- cient (mcc) over f1 score and accuracy in binary classification evaluation. BMC Genom. 21 (1), 1–13 (2020).
Carvalho, D. V., Pereira, E. M. & Cardoso, J. S. Machine learning interpretability: A survey on methods and metrics. Electronics 8 (8), 832 (2019).
Chou, J.-S. & Nguyen, T.-K. Forward forecast of stock price using sliding- window metaheuristic-optimized machine-learning regression. IEEE Trans. Ind. Inform. 14 (7), 3132–3142 (2018).
Vafeiadis, T., Diamantaras, K. I., Sarigiannidis, G. & Chatzisavvas, K. C. A com- parison of machine learning techniques for customer churn prediction. Simul. Model. Pract. Theory 55 , 1–9 (2015).
Ismail, M. R., Awang, M. K., Rahman, M. N. A. & Makhtar, M. A multi-layer percep- tron approach for customer churn prediction. Int. J. Multimed. Ubiquitous Eng. 10 (7), 213–222 (2015).
Riedmiller, M., Lernen, A. Multilayer perceptron. Machine learning lab special lecture, University of Freiburg 24 (2014).
Jain, H., Khunteta, A. & Srivastava, S. Churn prediction in telecommunication using logistic regression and logit boost. Procedia Comput. Sci. 167 , 101–112 (2020).
Pyecharts, https://github.com/pyecharts/pyecharts .
Download references
Thanks for Authors Contributions: “Conceptualization, Chenggang He; methodology, Chenggang He; validation, Chenggang He,Chris H.Q.Ding; investigation, Chenggang He; writing—original draft preparation, Chenggang He; writing—review and editing, Chenggang He; supervision, Chris H.Q.Ding.
This research was funded by the Scientific Research Foundation for High- level Talents of Anhui University of Science and Technology(2023yjrc120), Anhui Quality Engineering Project(2023cyts013), NSFC Key Project of International (Regional) Cooperation and Exchanges (61860206004), Natural Science Foundation of China (61976004,61572030).
Authors and affiliations.
School of Public Safety and Emergency Management, Anhui University of Science and Technology, No.15 Fengxia Road, Hefei, 230041, Anhui, China
Chenggang He
School Department of Computer Science and Engineering, University of Texas at Arlington, 701 S. Nedderman Drive, Arlington, TX, 76019, USA
Chris H. Q. Ding
School of Computer Science and Technology, Anhui University, 111 Jiulong Road, Hefei, 230039, Anhui, China
Chenggang He & Chris H. Q. Ding
You can also search for this author in PubMed Google Scholar
Conceptualization, C.H.; methodology, C.H.; validation, C.H.; investigation, C.H.; writing—original draft preparation, C.H.; writing—review and editing, C.H., C. H. Q. D.; supervision, C. H. Q. D.. All authors reviewed the manuscript.
Correspondence to Chenggang He .
Competing interests.
The authors declare no competing interests.
Publisher's note.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ .
Reprints and permissions
Cite this article.
He, C., Ding, C.H.Q. A novel classification algorithm for customer churn prediction based on hybrid Ensemble-Fusion model. Sci Rep 14 , 20179 (2024). https://doi.org/10.1038/s41598-024-71168-x
Download citation
Received : 27 January 2024
Accepted : 26 August 2024
Published : 30 August 2024
DOI : https://doi.org/10.1038/s41598-024-71168-x
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
Sign up for the Nature Briefing: AI and Robotics newsletter — what matters in AI and robotics research, free to your inbox weekly.
Recognition and classification are among the most important applications of machine learning. This recognition process is used to identify objects and humans. In particular, it plays a major role in authentication processes by identifying features such as human eyes, fingerprints, and facial patterns. Among these features, facial recognition is an evolving technology used in smartphones, attendance systems in offices, and healthcare centers. Several research efforts have been conducted to perform facial recognition using machine learning and deep learning algorithms. These algorithms have performed well on faces without masks, but they have struggled with masked faces, as most facial features are hidden by the mask. Therefore, an improved algorithm is needed for performing facial recognition on faces with and without masks. Since the COVID-19 outbreak, research has been focused on using deep learning algorithms to identify masked faces. However, these algorithms were typically trained on faces both with and without masks. In this paper, we propose a facial recognition approach for recognizing faces with and without masks. The common regions of the face in both scenarios are identified by cropping the image. These cropped regions are then subjected to feature extraction using histogram properties, SURF, and SIFT features. The dominant features are identified using a swarm intelligence approach called Glowworm Swarm Optimization. These dominant features are then trained using a neural network with a regression function. Finally, the performance of the proposed method will be evaluated based on accuracy, sensitivity, and specificity and compared to existing approaches, such as SURF, with different variations for facial recognition with and without masks.
This is a preview of subscription content, log in via an institution to check access.
Subscribe and save.
Price includes VAT (Russian Federation)
Instant access to the full article PDF.
Rent this article via DeepDyve
Institutional subscriptions
The data used to support the findings of created new data set, this study is available from the corresponding author upon request.
Generalized Regression Neural Network
Adaptive Histogram Equalization
Histogram Modification Function
Scale-Invariant Feature Transform
Speeded-Up Robust Features
Unsupervised Speeded-Up Robust Features
Hybrid Glowworm Swarm Optimization
Glowworm Swarm Optimization
Hemathilaka, Susith & Aponso, Achala. (2021). An Analysis of Face Recognition under Face Mask Occlusions. 41–48. https://doi.org/10.5121/csit.2021.111804 .
Aware, Mrunal & Labade, Prasad & Tambe, Manish & Jagtap, Aniket & Beldar, Chinmay. (2021). Attendance Management System using Face-Recognition. International Journal of Scientific Research in Computer Science, Engineering and Information Technology. 336–341. https://doi.org/10.32628/CSEIT217370 .
Alexiou, Michail & Papadakis Ktistakis, Iosif & Goodman, Garrett. (2021). Towards a Masked Face Recognition Algorithm: A Novel Rule Based Hybrid Algorithm. 1–6. https://doi.org/10.1109/SEEDA-CECNSM53056.2021.9566244 .
Baskaran, Kaaviya & P, Baskaran & V, Rajaram & Kumaratharan, Narayanaswamy. (2020). IoT Based COVID Preventive System for Work Environment. 65–71. https://doi.org/10.1109/I-SMAC49090.2020.9243471 .
Arora M, Naithani S, Areeckal A (2022) A web-based application for face detection in real-time images and videos. J Phys: Conf Ser 2161:012071. https://doi.org/10.1088/1742-6596/2161/1/012071
Article Google Scholar
Jiang, Andy & Sun, Yu. (2023). A Mobile Application to Mark Attendance using a Combined Backend of the Firestore Database and Amazon AWS Services. 87–98. https://doi.org/10.5121/csit.2023.130108 .
Pichetjamroen, Sasakorn & Rattanalerdnusorn, Ekkachan & Vorakulpipat, Chalee & Pichetjamroen, Achara. (2021). Multi-Factor based Face Validation Attendance System with Contactless Design in Training Event. 637–640. https://doi.org/10.1109/ECTI-CON51831.2021.9454779 .
Asaju, Christine & Vadapalli, Hima. (2021). Affects Analysis: A Temporal Approach to Estimate Students’ Learning. 1–7. https://doi.org/10.1109/IMITEC52926.2021.9714657 .
Mundial, I.Q., ul Hassan, M., Tiwana, M.I., Qureshi, W.S., &Alanazi, E.A. (2020). Towards Facial Recognition Problem in COVID-19 Pandemic. 2020 4rd International Conference on Electrical, Telecommunication and Computer Engineering (ELTICOM), 210–214.
Nursetiawati S, Josua DP, Atmanto D, Oktaviani F, Fardani AL (2020) Science Education in the Family Environment with the Experimental Method of Facial Cosmetics Plant Fertilization in the Covid-19 Pandemic Era. JurnalPendidikan IPA Indonesia 9:561–573
Tseng, F., Cheng, Y., Wang, Y., & Suen, H. (2022). Real-time Facial Expression Recognition via Dense & Squeeze-and-Excitation Blocks. Human-centric Computing and Information Sciences, 12. https://doi.org/10.22967/HCIS.2022.12.039
Awan, S.A., Ali, S., Hussain, I.,., Hassan, B., & Ashfaq Ashraf, S.M. (2021). Proficient Masked Face Recognition Method Using Deep Learning Convolution Neural Network in Covid-19 Pandemic. International Journal of Circuits, Systems and Signal Processing.
Savchenko A, Savchenko LV, Makarov I (2022) Classifying Emotions and Engagement in Online Learning Based on a Single Facial Expression Recognition Neural Network. IEEE Trans Affect Comput 13:2132–2143
Bradford, B., Yesberg, J.A., Jackson, J., & Dawson, P.A. (2020). Live Facial Recognition: Trust and Legitimacy as Predictors of Public Support for Police Use of New Technology. The British Journal of Criminology.
Srinivas, M.A., Kambam, V.R., Pinnu, S.K., Edla, S., &Komera, A.K. (2022). Facial Expression Recognition based Restaurant Scoring System Using Deep Learning. Emperor Journal of Applied Scientific Research.
Rifaee, M.A., Al Rawajbeh, M., AlOkosh, B., &AbdelFattah, F. (2022). A New approach to Recognize Human Face Under Unconstrained Environment. International Journal of Advances in Soft Computing and its Applications.
Enadula, S.M., SriramEnadula, A., &Burri, R.D. (2021). Recognition of Student Emotions in an Online Education System. 2021 Fourth International Conference on Electrical, Computer and Communication Technologies (ICECCT), 1–4.
Recto, I.J., & Devaraj, M. (2022). Synthetic Occluded Masked Face Recognition using Convolutional Neural Networks. 2022 IEEE International Conference on Industry 4.0, Artificial Intelligence, and Communications Technology (IAICT), 124–129.
Sekhar JNC, Domathoti B, Santibanez Gonzalez EDR (2023) Prediction of Battery Remaining Useful Life Using Machine Learning Algorithms. Sustainability 15(21):15283. https://doi.org/10.3390/su152115283
Nurhopipah, A., Rifai Azziz, I., &Suhaman, J. (2022). Behind the Mask: Detection and Recognition Based-on Deep Learning. (Indonesian Journal of Computing and Cybernetics Systems). Vol.16, No.1, pp. 67–78. https://doi.org/10.22146/ijccs.72075
Birkle, J., Weber, R., Möller, K., & Wagner-Hartl *, V. (2022). Psychophysiological Parameters for Emotion recognition – Conception and first evaluation of a measurement environment. Intelligent Human Systems Integration (IHSI 2022). https://doi.org/10.54941/ahfe100958
Gupta, S., Kumar, P., &Tekchandani, R. (2022). Facial emotion recognition based real-time learner engagement detection system in online learning context using deep learning models. Multimedia Tools and Applications, 1 - 30. https://doi.org/10.1007/s11042-022-13558-9
Bhattacharya, Pronaya & Handa, Anand & Zuhair, Mohd. (2021). Face Detection System for Health Care Units Using Raspberry PI. https://doi.org/10.1109/ICIEM51511.2021.9445367 .
Mittal, Parangat & Pandey, Kartikeya & Tawani, Peeyush & Rohilla, Rajesh. (2021). CNN-based Person Recognition System for Masked Faces in a post-pandemic world. 1–6. https://doi.org/10.1109/INCET51464.2021.9456416 .
Hu Y, Jiang Z, Zhu K (2022) An Optimized CNN Model for Engagement Recognition in an E-Learning Environment. Appl Sci 12(16):8007. https://doi.org/10.3390/app12168007
Download references
Not Applicable.
Author declared that no funding was received for this Research and Publication.
Authors and affiliations.
Dept of CSE-AIML, Geethanjali College of Engineering and Technology, JNTUK, Kakinada, Hyderabad, 501301, India
Kosuri Naresh Babu
Dept of IT, Velagapudi Ramakrishna Siddhartha Engineering College, Vijayawada, 520007, India
Suneetha Manne
You can also search for this author in PubMed Google Scholar
Correspondence to Kosuri Naresh Babu .
Ethical approval.
This article does not contain any studies with human participant and Animals performed by author.
The authors declare that there are no conflicts of interest regarding the publication of this paper.
Publisher's note.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
Reprints and permissions
Babu, K.N., Manne, S. A Novel Approach for Accurate Identification in Masked and Unmasked Scenarios using Glowworm Swarm Optimization and Neural Networks. Multimed Tools Appl (2024). https://doi.org/10.1007/s11042-024-20093-2
Download citation
Received : 15 December 2023
Revised : 09 August 2024
Accepted : 14 August 2024
Published : 31 August 2024
DOI : https://doi.org/10.1007/s11042-024-20093-2
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
Advertisement
You are accessing a machine-readable page. In order to be human-readable, please install an RSS reader.
All articles published by MDPI are made immediately available worldwide under an open access license. No special permission is required to reuse all or part of the article published by MDPI, including figures and tables. For articles published under an open access Creative Common CC BY license, any part of the article may be reused without permission provided that the original article is clearly cited. For more information, please refer to https://www.mdpi.com/openaccess .
Feature papers represent the most advanced research with significant potential for high impact in the field. A Feature Paper should be a substantial original Article that involves several techniques or approaches, provides an outlook for future research directions and describes possible research applications.
Feature papers are submitted upon individual invitation or recommendation by the scientific editors and must receive positive feedback from the reviewers.
Editor’s Choice articles are based on recommendations by the scientific editors of MDPI journals from around the world. Editors select a small number of articles recently published in the journal that they believe will be particularly interesting to readers, or important in the respective research area. The aim is to provide a snapshot of some of the most exciting work published in the various research areas of the journal.
Original Submission Date Received: .
Find support for a specific problem in the support section of our website.
Please let us know what you think of our products and services.
Visit our dedicated information section to learn more about MDPI.
Research on grid multi-source survey data sharing algorithm for cross-professional and cross-departmental operations collaboration.
2. grid engineering survey data and its characteristics, 3. research on grid survey data sharing algorithm, 3.1. survey data sharing methods combining differential privacy, 3.1.1. overview of the methodology, 3.1.2. discriminator feedback construction combining differential privacy.
Discriminator Weight Updates Combined with Differential Privacy |
discriminator weight T , discriminator loss function L , survey data data_x, generator synthesized shared data data_g, learning rate l_r, Privacy budget for differential privacy ep, Differential privacy sensitivity delta, first-order momentum estimation m, second-order momentum estimation v, threshold C, and Gaussian noise standard deviation S. m = 0 v = 0 sigma = sqrt (delta/(2×ep)) for each iteration in training://Each iterative step in the training process //Calculate the loss for real and generated data loss_real = LD(data_x,T ) loss_fake = LD(data_g,T ) //Gradient calculation, the gradient function is used to calculate the gradient of the loss function with respect to the model parameters grad_real = gradient(loss_real, T ) grad_fake = gradient(loss_fake, T ) //Merge the gradients and compute the average gradient grad = (grad_real + grad_fake)/2 //Updating the first- and second-order momentum estimates, beta1 and beta2 are the first- and second-order momentum parameters of Adam’s optimizer. m = beta1*m + (1−beta1)×grad v = beta2*v + (1-beta2)×(grad ) //Calculate the adaptive learning rate, t denotes the current number of iterations adaptive_lr = l_r×(sqrt(v/(1−beta2 ))) //Noise is added according to differential privacy requirements, and the normal_noise function generates noise based on a Gaussian distribution Noise = normal_noise(mean = 0, std = S) //Updating discriminator weights while considering privacy-preserving noise TD = TD−adaptive_lr×(m + noise) //Updating the privacy budget ep = ep−delta //Stop updating if the privacy budget is depleted or less than the threshold C if epsilon < 0 or epsilon < C: break return T |
Dynamic Noise Conditioning Algorithm |
Attenuation rate , Initial noise size , Survey data data_sources. noise_scales = {source: for source in data_sources} for source in data_sources://Iterate through each data source //Sample data from the current data source batch_data = sample_data(source) //Calculate the loss and gradient of the model on the current data loss = calculate_loss(model, batch_data) grad = calculate_gradient(loss, model.params) //Dynamically adjust the noise scale of the current data source according to the attenuation rate noise_scales [source] = noise_scales [source] × //Adding noise for differential privacy noise = normal_noise(mean = 0, std = noise_scales [source]) noisy_grad = grad + noise return noisy_grad//Gradient after output adaptive perturbation |
4. experiment, 4.1. experimental configuration and data sources, 4.2. experimental situation, 4.2.1. parameter settings, 4.2.2. evaluation index, 4.2.3. experimental results against survey data sharing methods combining differential privacy, comparison of algorithm performance with different number of sharers, comparison of algorithm performance under sharing between different professionals, comparison of algorithm performance under sharing between different departments, 4.2.4. experimental results of attribute encryption based permission change method, 5. conclusions, author contributions, data availability statement, conflicts of interest.
Click here to enlarge figure
Name | Content | Format | Structured vs. Unstructured | Real-Time vs. Non-Real-Time |
---|---|---|---|---|
Image data | Including remote sensing data, aerial data, laser point cloud data, etc. | TIFF, PNG, JPG, GeoTiff, IMG, GIF, BMP | Unstructured/ | real-time |
Sensors data | Includes pressure sensor data, radar sensor data, humidity sensor data, etc. | TXT, DAT, BRN, CSV | Structured, unstructured | real-time |
Basic control measurement data | Basic control measurement information element attribute information | XML, HTML, JSON, YAML, CSV | Structured | real-time |
Geotechnical data | Attribute information of exploration data elements of exploration points, etc. | XML, HTML, JSON, YAML, CSV | Structured | real-time |
3D modeling data | Three-dimensional modeling data of power grid engineering facilities and the surrounding environment | CGR, DWG, DXF, DWF, DGNPLN, RVT | Unstructured | non-real-time |
Model | ATLAS [ ] | DP-CGANS [ ] | DPGDAN [ ] | Our |
---|---|---|---|---|
LR | 0.7888 | 0.7303 | 0.7262 | 0.8547 |
SVM | 0.7762 | 0.7235 | 0.7061 | 0.8426 |
RF | 0.7748 | 0.7312 | 0.7133 | 0.8219 |
AVG | 0.7799 | 0.7283 | 0.7152 | 0.8397 |
The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
Zhang, J.; He, B.; Lv, J.; Zhao, C.; Yu, G.; Liu, D. Research on Grid Multi-Source Survey Data Sharing Algorithm for Cross-Professional and Cross-Departmental Operations Collaboration. Energies 2024 , 17 , 4380. https://doi.org/10.3390/en17174380
Zhang J, He B, Lv J, Zhao C, Yu G, Liu D. Research on Grid Multi-Source Survey Data Sharing Algorithm for Cross-Professional and Cross-Departmental Operations Collaboration. Energies . 2024; 17(17):4380. https://doi.org/10.3390/en17174380
Zhang, Jiyong, Bangzheng He, Jingguo Lv, Chunhui Zhao, Gao Yu, and Donghui Liu. 2024. "Research on Grid Multi-Source Survey Data Sharing Algorithm for Cross-Professional and Cross-Departmental Operations Collaboration" Energies 17, no. 17: 4380. https://doi.org/10.3390/en17174380
Further information, mdpi initiatives, follow mdpi.
Subscribe to receive issue release notifications and newsletters from MDPI journals
A not-for-profit organization, IEEE is the world's largest technical professional organization dedicated to advancing technology for the benefit of humanity. © Copyright 2024 IEEE - All rights reserved. Use of this web site signifies your agreement to the terms and conditions.
Deep learning techniques have revolutionized image classification by mimicking human cognition and automating complex decision-making processes. However, the deployment of AI systems in the wild, especially in high-security domains such as defence, is curbed by the lack of explainability of the model. To this end, eXplainable AI (XAI) is an emerging area of research that is intended to explore the unexplained hidden black box nature of deep neural networks. This paper explores the application of the eXplainable Artificial Intelligence (XAI) tool to interpret the underwater image classification results, one of the first works in the domain to the best of our knowledge. Our study delves into the realm of SONAR image classification using a custom dataset derived from diverse sources, including the Seabed Objects KLSG dataset, the camera SONAR dataset, the mine SONAR images dataset, and the SCTD dataset. An extensive analysis of transfer learning techniques for image classification using benchmark Convolutional Neural Network (CNN) architectures such as VGG16, ResNet50, InceptionV3, DenseNet121, etc. is carried out. On top of this classification model, a post-hoc XAI technique, viz. Local Interpretable Model-Agnostic Explanations (LIME) are incorporated to provide transparent justifications for the model's decisions by perturbing input data locally to see how predictions change. Furthermore, Submodular Picks LIME (SP-LIME) a version of LIME particular to images, that perturbs the image based on the submodular picks is also extensively studied. To this end, two submodular optimization algorithms i.e. Quickshift and Simple Linear Iterative Clustering (SLIC) are leveraged towards submodular picks. The extensive analysis of XAI techniques highlights interpretability of the results in a more human-compliant way, thus boosting our confidence and reliability.
Discover the world's research
IMAGES
VIDEO
COMMENTS
This survey paper provides a comprehensive overview of recent research and development in network security that uses graphs and graph-based data representation and analytics. The paper focuses on the graph-based representation of network traffic records and the application of graph-based analytics in intrusion detection and botnet detection. The paper aims to answer several questions related ...
Cryptography and network security should be narrower to include knowledge on safeguarding digital information and providing security services.
Network Security Monitoring (NSM) is a popular term to refer to the detection of security incidents by monitoring the network events. An NSM system is central for the security of current networks, given the escalation in sophistication of cyberwarfare. In this paper, we review the state-of-the-art in NSM, and derive a new taxonomy of the functionalities and modules in an NSM system. This ...
PDF | On Mar 26, 2019, Priya Chittibabu published A Comparative Analysis Of DES, AES and RSA Crypt Algorithms For Network Security in Cloud Computing | Find, read and cite all the research you ...
Security-by-design is a way to build a network where security is considered holistically in the whole network from its first concept, through the design, development, installation, configuration and maintenance of the network and to the finalisation of the useful life of the network.
Network security situational awareness is based on the extraction and analysis of big data, and by understanding these data to evaluate the current network security status and predict future development trends, provide feedback to decision-makers to make corresponding countermeasures, and achieve security protection for the network environment. This article focuses on artificial intelligence ...
Reliability in the context of network security refers to the ability of a system to consistently and dependably provide a high level of security across the network. A reliable system ensures that security measures are consistently applied and maintained, minimizing the risk of unauthorized access, data breaches, and other security incidents.
The paper explores applied cryptography concepts in information and network systems security to prevent cyberattacks and improve secure communications. The contribution of the paper is threefold: First, we consider the various cyberattacks on the different cryptography algorithms in symmetric, asymmetric, and hashing functions.
In this paper, we discuss AI-based protection techniques, according to a security life-cycle consisting of several phases: (i) Prepare; (ii) Monitor and Diagnose; and (iii) React, Recovery and Fix. For each phase, we discuss relevant AI techniques, initial approaches, and research directions.
According to the principles of security management, intuitive scientificity, and scalability, an information security system architecture based on representation and metric deep learning algorithms was designed. Two key algorithm technologies, interactive interfaces, neural network algorithms, and convolutional operation algorithms, were used for information security prediction analysis and ...
1 INTRODUCTION With the recent interest and progress in the development of internet and communication technologies over the last decade, network security has emerged as a vital research domain. It employs tools like firewall, antivirus software, and intrusion detection system (IDS) to ensure the security of the network and all its associated assets within a cyberspace. 1 Among these, network ...
Consequently, there is a lack of an in-depth review of WSN we find that there are two aspects to the survey research; one with minimal information about assaults and other studies that explore network security and its impact on energy dissipation, using our understanding of the security difficulties in WSN.
This paper discusses lightweight block ciphers, stream ciphers, and hybrid ciphers. The report evaluates security algorithms, comparing performance and robustness with the computational complexity of these techniques. Finally, the survey presents IoT security challenges, threats, and attacks with their mitigation techniques.
To answer the fourth and final research question of this paper (RQ4), the literature relevant to our research questions was scrutinized to highlight potential research gaps and identify opportunities for future AI for cybersecurity research.
First step of getting the help from algorithm is to be studied and compared their parameters. This paper presents a review that comparative study of algorithms taken by many authors.
Effectively, we have chosen the research papers from the year 2011 to 2020, which are based on cybersecurity issues with deep learning concepts. Ultimately, we analyzed 80 research papers from different kinds of journals and the deeply analyzed survey are effectively mentioned in the below section.
Abstract Today's world is highly network interconnected owing to the pervasiveness of small personal devices (e.g., smartphones) as well as large computing devices or services (e.g., cloud computing or online banking), and thereby each passing minute millions of data bytes are being generated, processed, exchanged, shared, and utilized to yield outcomes in specific applications. Thus ...
However, the proposed algorithm to be implemented in this research paper will lower the level of possible attacks and enhancing security without compromising the performance of the network and also improving the power consumption within WSNs in IoT.
The outcomes of this paper provide valuable directions for further research and applications in the field of applying effective and efficient intelligent techniques in network analytics. This article is organized into four sections. The first section provides an introduction and background to the research area.
With the rapid development of network communication technology, the continuous deepening of Internet application and the increasing enrichment of information, the Internet has become an important infrastructure of human society. With the development of network technology and the increasing complexity of network topology, the supervision of network is facing great challenges. Among them ...
Network risk assessment should include the impact of the relationship between vulnerabilities, in order to conduct a more in-depth and comprehensive assessment of vulnerabilities and network-related risks. However, the impact of extracting the relationship between vulnerabilities mainly relies on manual processes, which are subjective and inefficient. To address these issues, this paper ...
In this paper, the graph sample and aggregate-attention network with war strategy optimization algorithm for cyber security in the 5G wireless communication network (CS-5GWCN-GSAAN-WSOA) is proposed in 5G mobile networks to identify cyber threats. Initially, the input data are amassed from the 5G-NIDD dataset.
The SEI Digital Library provides access to more than 6,000 documents from three decades of research into best practices in software engineering. These documents include technical reports, presentations, webcasts, podcasts and other materials searchable by user-supplied keywords and organized by topic, publication type, publication year, and author.
In order to better carry out the research on customer churn rate, this paper focuses on the theoretical basis of the Support Vector Machine algorithm, Random Forest algorithm, K-neighborhood ...
Network security consists of the provisions and policies adopted by a network administrator to prevent and monitor unauthorized access, misuse,... | Explore the latest full-text research PDFs ...
In [] research, researchers suggested a mask detection technique that relies on HOG (Histogram of Gradient) attributes classifier and SVM (Support Vector Machine) to establish whether the face is masked or not.The proposed method was tested using over 10,000 randomly selected images from the Masked Face-Net database, and it correctly classified 98.73% of the tested images.
This paper addresses the problem of multi-source survey data sharing in power system engineering by proposing two improved methods: a survey data sharing method combined with differential privacy and a permission change method based on attribute encryption. The survey data sharing method integrated with differential privacy achieves effective cross-professional and cross-departmental data ...
This paper addresses the emerging problems and challenges related to security in WSNs. The proposed solution, namely the DSD-RSA algorithm, integrates existing models such as the S-MAC protocol, AES algorithm, and RSA algorithm. DSD-RSA aims to minimize power consumption and end-to-end delay while maximizing network throughput in WSNs.
Deep learning techniques have revolutionized image classification by mimicking human cognition and automating complex decision-making processes. However, the deployment of AI systems in the wild, especially in high-security domains such as defence, is curbed by the lack of explainability of the model. To this end, eXplainable AI (XAI) is an emerging area of research that is intended to explore ...
Cryptography is the encryption and decryption of data with secret keys using various algorithms. In this paper network security are described on the basis of the services of security.