A Detailed Review for Marketing Decision Making Support System in a Customer Churn Prediction

INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 9, ISSUE 04, APRIL 2020 ISSN 2277-8616

A Detailed Review For Marketing Decision Making Support System In A Customer Churn Prediction

Mr. M. John Britto, Dr. R. Gobinath

Abstract:Churn prediction is one of the important issues in customer relationship management (CRM). It has become increasingly important to retain existing customer than acquiring new one. This paper presents a review of customer’s churn prediction in various domains like telecommunications, retail banking, e-banking, energy sectors, insurance, and so on. The study shows a huge number of attributes that were used to develop customer churn model by many researchers. It reveals various techniques that are used till date in churn prediction. Modelling techniques such as Logistic Regression, Neural Network Model, Random Forest, Decision Tree, Support Vector Machine and Rough set approach are implemented for the churn detection. The findings reveal that the customer churn prediction through predictive analytics will fetch more accurate outcomes while comparing with other similar approaches in prediction. There is large research scope in Customer churn prediction and Customer retention through predictive analytics.

Keywords: CRM, Churn, Predictive Analytics, Data Mining, Attrition. ————————————————————

1 INTRODUCTION Hence it is the task of Bank Credit Card Account The digital media is one of the most favourite and powerful Management System (BCCAMS) to preserve the existing media that uses large amount of data in database which customers. Through highly efficient data mining systems faces a problem of irrelevancy and some major problems in and predictive analytics it is possible to predict churn out of transactions. There is an essential need of large data particular bank based on various available attributes repositories to store and manage these types of data. The collected from the past history of old customers. Data immense challenge for such data transactions are due to mining methods such as decision tree, naive bayes, the irrelevancy in data as well as due to missing values, random forest, artificial neural networks, inductive rule which makes difficulty in extracting necessary information. learning, and support vector machine to find out the Furthermore, with digital transaction of the banking industry, churn[1,3– 8].All these techniques are implemented not it is necessary for banks to keep customers in electronic only in banking but also in medical systems, insurance, channels like internet or mobile applications. Reason for telecommunication, gaming, automobile industries, retail this is that digital customer is more profitable than the marketing etc.,[12]–[15]. Previously customer retention traditional transaction customer. There is no need for a technique in credit card churn prediction was done using physical branch or a station for client service supervised techniques. Later the research work in churn communication in transactions. Customer attrition in started to use unsupervised techniques[4,16,18]. But it banking industry or electronic banking may lead to could not end with better results. So, through hybrid economic crisis and makes the situation worse. It is Predictive analytics techniques we might bring better essential to predict the customer churn earlier and retaining accuracy. methods should be implemented. In recent years there has been huge increase in the amount of data that is collected Data Mining and processed to extract meaningful and valuable Data mining is an important component of every CRM information across wide areas of business. The meaningful framework that facilitates analysis of business problems, information that is obtained is then used by the companies prepare data requirements, and build, validate and evaluate for CRM [1]–[4]. Although there are many techniques that models for business problems [2], [3], [9]. The data mining have been effectively applied in predicting customer churn process and algorithms enable firms to search, discover like using SVM[5], [6], logistic regression [7], decision hidden patterns and correlations among data, and to extract Trees [7], Naive Bayes and neural networks [4][8]–[10] in relevant knowledge buried in commercial data warehouses, the domain of airlines, banking, energy sector, in order to gain broader understanding of business. Data telecommunication, retail banking and many more sectors, mining uses sophisticated statistical data search algorithms deep learning, predictive analytics approaches for this to find, discover hidden patterns and relationships for circumstances still have lots to be explored. However every extracting knowledge buried in corporate data customer can have more than one credit card more than warehouses,[12] or information that visitors have dropped one bank so there might be lot of chances for a customer to about their experience, most of which can lead to churn out of particular bank [11]. improvements in the understanding and use of the data in order to detect significant patterns and rules underlying consumer’s behaviours.

______Related Work Nowadays, there exists a plethora of machine learning  Mr. M. John Britto1, Dr. R. Gobinath2 approaches to customer data mining and retention  Research Scholar, Department of BCA/IT, VISTAS, Chennai, modelling. It is ranging from classical regression to neural Tamil Nadu, India. Associate Professor, Department of Computer networks to random forests (e.g., see [12, 17, 18, and 20] Science, VISTAS, Chennai, Tamil Nadu, India. for a general topic overview). The experiments in [10], [19], [20] showed that neural networks typically outperform

3698 IJSTR©2020 www.ijstr.org INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 9, ISSUE 04, APRIL 2020 ISSN 2277-8616 logistic regression and decision trees in churn prediction. could target only those customers instead of all customers Applications of social network analysis to customer for giving some incentives to retain them. retention, however, are often limited due to poor availability of data on customer peer networks[13]. Most progress in Customer Churn Prediction this direction has been achieved in telecommunication The churn means those customers who will leave in near industry, where social networks are naturally observed from future. There is essential need to predict those customers the call and message records (e.g., see [1, 2, 13, 17, and on behalf of some parameter to initiate some suitable action 20]). Constructing networks of bank customers requires to minimize their leaving. The most of the mobile phone additional steps, such as targeted surveys [3], mining the companies invest under CRM (customer relationship databases of customers and their transactions [19], and, management) technology. potentially, employing big data approaches[15], [21]–[23] for harnessing customer information from different sources, Types of Churners including online social media [22].Deep learning (DL) Churners are classified into two main categories that are methods continue to attract increasing interest in customer voluntary and involuntary[25]. The voluntary churners are churn prediction, while being a relatively new tool in further more subdivided into deliberate and incidental customer analytics. DL concepts into customer churn churners. Involuntary churners are those customers prediction and retention models in retail banking with multi- removed from the list of customers due to their non- layer feed-forward architecture, can effectively capture payment, fraud and so on. The voluntary churners are features of the underlying customer data [24]. difficult to find due to that customers want to terminate from his services providers. The incidental churners come due to CRM (Customer Relationship Management) incident because the churners have no plan to leave but The CRM covers the Cross-Selling, Up-Selling, Customer this one done due to some reasons like change of location, Retention, New Customer Acquiring module. The main change in financial position etc. focus of Customer retention is to retain current customer in an organization. The following are the reasons There is rare Modelling Techniques of Churn Prediction or no chance of the new customer in the telecommunication It is the basic need of the companies to develop an efficient and banking Industries due to saturation. Acquiring of new and effective model to manage customers churn. There are customer is costly for a company due to various reasons. so many modelling techniques that are used to predict There is ten times increase in expenditure when acquiring a customers churn in different organization. Here is the table new customer related to the expenses of retaining the of customer churn prediction models/techniques and current existing customer[25].Churn prediction helps to various algorithms. identify expected churning customer, so that the companies

Table 1: Various Algorithms for Churn Prediction.

ALGORITHMS/ YEAR AUTHOR TITLE DOMAIN FEATURES METHODOLOGIES Improved credit card Precision, sensitivity, churn prediction based R. RajaMohamed, J. Credit card (e- Modified rough k-means specification, 2017 on rough clustering and Manoharan banking) algorithm accuracy and supervised learning misclassification techniques Gauging and A frame work with Neural foreseeing customer Portuguese retail Network approach. CRISP methodology 2018 Nelson Rosa churn in banking bank And ANN. industry- A Neural CRISP methodology Network Approach EDFFNN – Enhanced Enhanced deep feed Deep Feed Forward ROC Curve, F1 Sandeep Kumar forward neural network Neural Network. Score, Recall, 2019 Hegde, Monica R model for the customer Banking Precision and Mundada attrition analysis in confusion matrix banking sector

Fatemeh Safinejad, A fuzzy dynamic model Elham Akhond for customer churn LRFM is used as 2018 Retail Banking Fuzzy dynamic model Zadeh, Behrouz H. prediction in retail attributes. Far banking Improved churn prediction based on J. Vijaya, E. Hybrid supervised and Sensitivity, 2018 Supervised and Telecommunication Sivasankar unsupervised techniques specificity, accuracy unsupervised hybrid data mining system Precision, recall, Predicting customer Sanjay Kumar, Multi layered ANN consists accuracy are the 2019 churn using Artificial Telecommunication Manish Kumar with 3 dense layers evaluation measures Neural Network used. Arash Barfar, Balaji Applying Behavioral Behavioral economics Behavioral 2017 Padmanabhan, et. economics in predictive B2B concepts economics, 3699 IJSTR©2020 www.ijstr.org INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 9, ISSUE 04, APRIL 2020 ISSN 2277-8616

Al. analytics for B2B Theoretical service churn: finding from quality, Tangibles, service quality data Reliability, Responsiveness, Assurance,

ALGORITHMS/ YEAR AUTHOR TITLE DOMAIN FEATURES METHODOLOGIES Early prediction of Sandeep Yadav, HR Dataset, salary, employee attrition Hr Attrition dataset Logistic Regression, SVM, 2018 Aman Jain, Deepti promotion, using data mining from Kaggle website Random Forest, Singh satisfaction level. techniques Solving imbalance Erdem Kaya, Xiowen Behavioral attributes Credit card classification, spatio- Dong, Yoshihiko 2018 and financial churn sampling of OECD SVM –SMOTE temporal choice Suhara, Selim prediction countries patterns are taken Balcisoy, Alex Sandy into account Churn and non-churn Ramakanth of customers in Used supervised Portuguese banking ELM – Extreme Learning 2018 Mohanthy, C. Naga banking sector using and unsupervised sector Machine Ratna Sree Extreme Learning learning Machine Transfer learning and Uzair Ahmed, meta classification ROC is used for Telecommunication. 2017 Asifullah Khan, Yeon based deep churn TL-DeepE performance Cell2cell, Orange. Soo Lee prediction for telecom measures. industry Customer churn Abdelrahim Kasem prediction in telecom Telecommunication- 2019 Ahamed, Assef Jafer, XGBOOST algorithm IMEI data are used using machine learning Syria Tel. KadanAljoumaa in Big Data platform Anderi Simion Balanced between Constantineesu, Deep Neural pipeline Pharmaceutical 2018 Deep Neural pipeline High recall rate and Andrei Ionut Damian, for churn prediction industry precision rate Nicolae Tapus

Here is a list of those attributes that affect churning process The decision tree is the most prominent predictive model and help us to predict the churners. Customer that is used for the purpose of classification of upcoming demographics data, Customer data, Complaint data, Bill trial. The decision tree consists of two steps, tree building info, Payment info, Customer age, Fault report, Payment and tree pruning. In tree building the training set data is type, Consumption level rates, Area of customer, Quality of recursively partitioned in accordance with the values of the services, Purchases history, Survey report, and attributes[28]. This process goes on until there is no one Demographic details. The techniques that are used to partition is left to have identical values. During this process predict the Customer attrition are as follows. noisy data and outliers are removed. The largest estimated error rate branches are selected and then removed in Artificial Neural Networks (ANN) pruning. Pruning technique is used to reduce the size of a The Neural Networks Model (NNM) is used to elaborate tree and thus complexity of the decision is reduced which in functionality like non-linear. The model holds the capability turns reflects on accuracy of customer churn prediction. to learn due to its comparable data processing structure. These techniques provide successful results after applying Support Vector Machine (SVM) on many problems like classification [9][26]. The model is The SVM classifier deals with linear permutation of subset dissimilar to classification model as well as decision tree of the training set by finding a maximum edge over due to its likely hood prediction. The neural network has energized plane. The SVM plots the data into high several techniques having merits and demerits. The dimensional features space closing to infinite with the help researcher suggests deep neural network is better than of most important part if vectors are nonlinearly divisible decision tree and regression analysis model of churn input features and then categorize the data by the highest prediction [27]. scope hyper-plane

Linear Regression Model (LRM) Fuzzy Logic Algorithm To predict customer satisfaction and customer churn the A fuzzy logic technique is very simple to understand due to regression analysis model technique can also be its very simple mathematical concepts and fuzzy reasons. implemented which is a supervised learning model. In this Fuzzy logic has the property of flexibility, tolerant of model a data set of past observation is used to see future indefinite data. The function of random data can be values of explanatory and numerical targeted variable. implemented in this model. Evolutionary Learning Data Mining Techniques (DMEL) Data mining by evolutionary Decision Tree (DT) learning techniques is inherited classification technique. Such type of genetic algorithms has some set of rules for

DMEL technique[29]. The DMEL applies these rules on function is achieved. In k mean clustering, the most some given dataset that provide decision making results. important point is to find the numbers of clusters that is K-means clustering optimum as well as the distance between cluster mean and In K mean cluster in approach, in the first step we select k objects. The algorithm works until no new cluster element objects that have their centre (mean). In this method the leave a cluster and enter into other cluster and no new remaining objects are not selected yet are assigned to centre point is set for any cluster. When this target is cluster with respect to the similarity of the object with achieved the algorithm is stopped. Here is the table of cluster. These similarities are measured on the behalf of the customer churn prediction models/techniques based on distance between cluster mean and object to get the new behavioral economics and Deep Learning. centre point. These steps are repeated until the required

Table2: Behavioral economics, Deep Learning concepts for customer churn prediction. Methodologies/Te S.no Author Topic Description Future scope chniques used Behavioral economics, Theoretical service Applying Behavioral Behavioral Arash Barfar, quality, Tangibles, economics in predictive economics Both behavioral and Balaji Reliability, analytics for B2B churn: concepts Technical methodologies Padmanabhan, Responsiveness, finding from service quality can be implemented in et. Al [2017] Assurance, data Churn prediction models. Empathy[30]

Methodologies/Te S.no Author Topic Description Future scope chniques used Behavioral Features might Erdem Kaya, Solving imbalance be combined with other Xiowen Dong, classification, spatio- advanced characteristics Yoshihiko Behavioral attributes and SVM –SMOTE temporal choice such as customers’ Suhara, Selim financial churn prediction patterns are taken into purchase and search Balcisoy, Alex account history and feedbacks in Sandy[2018] social media. Behavioral economics can Three deep neural be taken into account with Automated Feature network architecture Three deep these three deep neural V.Umayaparvathi Selection and Churn were created. It neural network network architecture for , K. Iyakutti[2017] Prediction using Deep eliminates manual Learning Models. predicting domain Learning Models. feature engineering independent customer process churn It is five layer Deep EDFFNN – neural Network model Enhanced Deep with six nodes in each Enhanced deep feed Feed Forward layer. It has been Sandeep Kumar It can be implemented with forward neural network Neural Network, implemented using Hegde, Monica R other parameters with model for the customer Tukey outlier optimized data pre- Mundada [2019] rough set approach for attrition analysis in banking algorithm processing, data churn prediction sector exploration, feature scaling, and Adam optimizer algorithms for churn prediction Integration of information from Uzair Ahmed, Transfer learning and meta Transfer learning with Unsupervised learning Asifullah Khan, classification based deep TL-DeepE GP-Adaboost technique can be used for Yeon Soo churn prediction for ensemble improves customer churn prediction. Lee[2017] telecom industry churn prediction accuracy Anderi Simion Balanced between Behavioral economic Constantineesu, High recall rate and Deep Neural pipeline for Deep Neural concepts can be used for Andrei Ionut precision rate used to churn prediction pipeline independent domain Damian, Nicolae find accuracy in finding customer churn prediction Tapus [2018] churning customers.

predicting customer churn[30]. It fills the gap between Behavioral Economics domain specific pattern and Behavioral pattern of customer. Customers’ spatio-temporal behaviour can be considered Behavioral economics can inspire feature engineering and for churn prediction. Applying Behavioral economics will helps to build parsimonious models for domain independent fetch better results in predicting customer attrition and also customer churn prediction. helps to predict domain independent churn prediction. The dynamic spatio-temporal Behavioral features such as diversity, loyalty, and regularity can be included in 3701 IJSTR©2020 www.ijstr.org INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 9, ISSUE 04, APRIL 2020 ISSN 2277-8616

CONCLUSION [9]. S. H. Dolatabadi and F. Keynia, ―Designing of The churn prediction done in recent years are mostly customer and employee churn prediction model concentrated on unsupervised learning with few based on data mining method and neural predictor,‖ measurements which makes an effective way of extracting 2nd Int. Conf. Comput. Commun. Syst. ICCCS useful information in CRM. There are various ways to 2017, pp. 74–77, 2017, doi: predict churns in a customer system. Behavioral economics 10.1109/CCOMS.2017.8075270. of customers can be used for domain independent [10]. N. Belém, ―Gauging and Foreseeing Customer customer churn prediction. Deep learning techniques are Churn in the Banking Industry : A Neural Network also implemented for predicting customer churn in various Approach,‖ 2018. areas such as behavioural economics, demographic, [11]. R. Rajamohamed and J. Manokaran, ―Improved transaction, segmentation. Many researchers concentrate credit card churn prediction based on rough on developing solution for customer churn detection using clustering and supervised learning techniques,‖ forecast model with various large number of attributes Cluster Comput., vol. 21, no. 1, pp. 65–77, 2018, which includes segmentation, account information, payment doi: 10.1007/s10586-017-0933-1. information, complaint information, and services [12]. A. O. Oyeniyi and A. B. Adeyemo, ―Customer Churn information. This paper examines various exhibiting Analysis In Banking Sector Using Data Mining techniques such as LR, NNM, DT, FL, SVM and DME for Techniques,‖ African J. Comput. ICT, vol. 8, no. 3, the domain based churning detection purpose in different pp. 165–174, 2015. domains such as online gaming, online retail banking, [13]. B. H. Far, Applications of Data Management and insurance, energy sector, and telecommunication and so Analysis Case Studies in Social Networks. . on. A framework can be examined for domain independent [14]. A. P. Patil, M. P. Deepshika, S. Mittal, S. Shetty, S. customer churn prediction using deep neural network S. Hiremath, and Y. E. Patil, ―Customer churn techniques. prediction for retail business,‖ 2017 Int. Conf. Energy, Commun. Data Anal. Soft Comput. ICECDS REFERENCE 2017, pp. 845–851, 2018, doi: [1]. K. Coussement, S. Lessmann, and G. Verstraeten, 10.1109/ICECDS.2017.8389557. A comparative analysis of data preparation [15]. F. Shirazi and M. Mohammadi, ―A big data analytics algorithms for customer churn prediction: A case model for customer churn prediction in the retiree study in the telecommunication industry, vol. 95. segment,‖ Int. J. Inf. Manage., vol. 48, no. February, Elsevier B.V., 2017. pp. 238–253, 2019, doi: [2]. S. Yadav, A. Jain, and D. Singh, ―Early Prediction of 10.1016/j.ijinfomgt.2018.10.005. Employee Attrition using Data Mining Techniques,‖ [16]. U. Ahmed, A. Khan, S. H. Khan, A. Basit, I. U. Haq, Proc. 8th Int. Adv. Comput. Conf. IACC 2018, pp. and Y. S. Lee, ―Transfer Learning and Meta 349–354, 2018, doi: 10.1109/IADCC.2018.8692137. Classification Based Deep Churn Prediction System [3]. J. Moeyersoms and D. Martens, ―Including high- for Telecom Industry,‖ pp. 1–10, 2019. cardinality attributes in predictive models: A case [17]. Y. Li and B. Wang, ―A study on customer churn of study in churn prediction in the energy sector,‖ commercial banks based on learning from label Decis. Support Syst., vol. 72, pp. 72–81, 2015, doi: proportions,‖ IEEE Int. Conf. Data Min. Work. 10.1016/j.dss.2015.02.007. ICDMW, vol. 2018-Novem, pp. 1241–1247, 2019, [4]. M. Shan, ―Building Customer Churn Prediction doi: 10.1109/ICDMW.2018.00177. Models in Fitness Industry with Machine Learning [18]. J. Vijaya and E. Sivasankar, ―Improved churn Methods,‖ 2017. prediction based on supervised and unsupervised [5]. A. K. Ahmad, A. Jafar, and K. Aljoumaa, ―Customer hybrid data mining system,‖ Lect. Notes Networks churn prediction in telecom using machine learning Syst., vol. 9, pp. 485–499, 2018, doi: 10.1007/978- in big data platform,‖ J. Big Data, vol. 6, no. 1, 2019, 981-10-3932-4_51. doi: 10.1186/s40537-019-0191-6. [19]. Y. Chen, Y. R. G. B, V. Lyubchich, and T. Winship, [6]. N. Gordini and V. Veglio, ―Customers churn Pakdd 2018, vol. 2, no. February. Springer prediction and marketing retention strategies. An International Publishing, 2018. application of support vector machines based on the [20]. S. hegde and M. R. Mundada, ―Enhanced Deep AUC parameter-selection technique in B2B e- Feed Forward Neural Network Model for the commerce industry,‖ Ind. Mark. Manag., vol. 62, pp. Customer Attrition Analysis in Banking Sector,‖ Int. 100–107, 2017, doi: J. Intell. Syst. Appl., vol. 11, no. 7, pp. 10–19, 2019, 10.1016/j.indmarman.2016.08.003. doi: 10.5815/ijisa.2019.07.02. [7]. Xiaojun Ma, Shengjun Zhai, Yingxian Fu, Leonard [21]. A. Amin et al., ―Customer churn prediction in the Yoonjae Lee, and Jingxuan Shen, ―Predicting the telecommunication sector using a rough set Occurrence and Causes of Employee Turnover with approach,‖ Neurocomputing, vol. 237, pp. 242–254, Machine Learning,‖ Comput. Eng. Appl., vol. 8, no. 2017, doi: 10.1016/j.neucom.2016.12.009. 3, pp. 217–227, 2019. [22]. A. Haldorai, A. Ramu, S. Mohanram, C. Chee, and [8]. A. De Caigny, K. Coussement, K. W. De Bock, and O. Editors, EAI International Conference on Big S. Lessmann, ―Incorporating textual information in Data Innovation for Sustainable Cognitive customer churn prediction models based on a Computing EAI/Springer Innovations in convolutional neural network,‖ Int. J. Forecast., no. Communication and Computing. 2018. xxxx, 2019, doi: 10.1016/j.ijforecast.2019.03.029. [23]. H. Sayed, M. A. Abdel-Fattah, and S. Kholief, 3702 IJSTR©2020 www.ijstr.org INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 9, ISSUE 04, APRIL 2020 ISSN 2277-8616

―Predicting potential banking customer churn using Apache Spark ML and MLlib packages: A comparative study,‖ Int. J. Adv. Comput. Sci. Appl., vol. 9, no. 11, pp. 674–677, 2018, doi: 10.14569/ijacsa.2018.091196. [24]. A. Predescu, A Multiple-Layer Clustering Method for Real-Time Decision Support. Springer International Publishing, 2019. [25]. M. Hassouna, A. Tarhini, T. Elyas, and M. S. Abou Trab, ―Customer Churn in Mobile Markets: A Comparison of Techniques,‖ Int. Bus. Res., vol. 8, no. 6, pp. 224–237, 2015, doi: 10.5539/ibr.v8n6p224. [26]. S. Yao, Signal2Vec : Time Series Embedding, no. December 2016. Springer International Publishing, 2019. [27]. A. Simion-Constantinescu, A. I. Damian, N. Ţǎpuş, L. G. Piciu, A. Purdilǎ, and B. Dumitrescu, ―Deep Neural Pipeline for Churn Prediction,‖ Proc. - 17th RoEduNet IEEE Int. Conf. Netw. Educ. Res. RoEduNet 2018, 2018, doi: 10.1109/ROEDUNET.2018.8514153. [28]. A. J. Nathan and A. Scobell, ―How China sees America,‖ Foreign Affairs, 2012. [Online]. Available: https://github.com/HumayunTiwana/Churn- Prediction/blob/master/Capstone_Churn_Prediction. Rmd. [29]. W. H. Au, C. C. Chan, and X. Yao, ―A novel evolutionary data mining algorithm with applications to churn prediction,‖ IEEE Trans. Evol. Comput., vol. 7, no. 6, pp. 532–544, 2003, doi: 10.1109/TEVC.2003.819264. [30]. A. Barfar, B. Padmanabhan, and A. Hevner, ―Applying behavioral economics in predictive analytics for B2B churn: Findings from service quality data,‖ Decis. Support Syst., vol. 101, pp. 115–127, 2017, doi: 10.1016/j.dss.2017.06.006.