PREDICTING CUSTOMER USING SVM

ABSTRACT

Support vector machines (SVMs) are promising methods for the prediction of ecommerce

customer churning and profitability because SVMs use a risk minimization principal that consists

of the empirical error and the regularized term predicting the switching probability of an insured

to other ecommerce. In addition, this study examines the feasibility of applying SVM in

ecommerce customer churning and profitability by comparing it with other methods such as

artificial neural network (ANN) and logit model. This study proves that SVM provides a

promising alternative to predict customer churning and profitability in ecommerce.

1
 CHAPTER ONE

1.0 INTRODUCTION

Data classification is the process of sorting and categorizing data into various types or any other

distinct class. Data classification enables the separation and classification of data according to

data set requirements for various business objectives (Vapnik, 1995). It is mainly a data

management process. Classification in data mining consists of predicting a particular outcome

based on the given input (Guyon,2002). In order to predict the outcome, the algorithm processes

a training set containing a set of attributes and the respective outcome, usually called goal or

prediction attribute (Kim,2003). The algorithm tries to find relationships between the attributes

that would make it possible to predict the outcome. Next the algorithm is given a data set not

seen before, which is called as prediction set, that contains the same set of attributes, except for

the prediction attribute – not yet known. The algorithm analyses the input and produces a

prediction. The prediction accuracy defines how “good” the algorithm is (Ben-David, 1995).

Classification models predict categorical class labels; and prediction models predict continuous

valued functions (Y. Ken, 2010). For example, we can build a model to classify bank loan

applications as either safe or risky, or a prediction model to predict the expenditures in dollars of

potential customers on computer equipment given their income and occupation.

Data mining tools predict patterns and behaviors, allowing businesses to effect knowledge-

driven decisions (Chowdhury, 2011). The automated, prospective analyses offered by data

mining tools move beyond the analysis of past events provided by retrospective tools typical

of decision support system.

2
Customer churn is a marketing-related term which means customers defect to another supplier

or purchase less (Azamathulla, 2011). As existing customers are an important source of

business profits, being able to identify customers who show signs that they are about to leave

can create more income for businesses. This is especially more important for online

customers, as the phenomenon of customer churn appears to be very rapid and difficult to

grasp. If companies do not take measures to hold customers before their status deteriorates, the

customers may never come back, resulting in wasted investment and loss of income. A timely

retention strategy can keep customers, and it is the best way to retain customers (Vapnik,

1998).

1.1 THE E-COMMERCE BUSINESS MODEL

E-commerce (David, 2000) (electronic commerce or EC) include the buying and selling of goods

and services and the transmitting of funds or data, over an electronic network. These business

transactions occur between business-to-business (B2B), business-to-consumer (B2C), consumer

to-consumer (C2C) or consumer-to-business (C2B). The terms e-commerce and e-retailer are

often used interchangeably in this work. We are primarily interested in e-retail business which is

a form of electronic commerce that allows consumers to directly buy goods or services from a

seller over the Internet using a web browser or a mobile app. It is projected that in the year 2017

the online e-retail industry will grow upwards of 600 billion dollars. Some of the household e-

retail names are Amazon, Alibaba, Walmart, e-bay, Staples, Macy’s, Apple, Konga, Jumia etc.

While most of these e-retailers operate on a B2C business model, a B2B model or a combination

of both is also common.

3
Many businesses have migrated from owning a Brick mortar shop alone to include e-retail

business to cater to the needs of the customer and to keep up with the competition while others

like Amazon, Alibaba follow only the e-commerce route.

Customer loyalty (Srini, 2002) is an important driver to many E-retailers as the cost of acquiring a

new customer is a significant effort in comparison to the cost of retaining one. Unlike a brick and

mortar shopping experience that involves a look and feel, location advantage and human

interaction component among others, the e-retail business model comes packaged in a single

website from the landing page to exit. Therefore it is the most important priority of these

companies to entice the customer with great line of products, pricing, attractive offers,

recommendations, personalization, etc to create a desirable shopping experience.

1.2 STATEMENT OF PROBLEM

Over the next few years, the industry's biggest marketing challenge will be to control churn rates
by identifying those customers who are most likely to leave and then taking appropriate steps to
retain them. The first step therefore is predicting churn and profitability likelihood at the
customer level. The Customer Churn and profitability Prediction problem has two major
characteristics:

 The first is that the number of churn customers (the negative examples) is small
(2% in the total examples);

 The second is accuracy. Consequently, for a carrier with 1.5 million subscribers,
improving the monthly prediction accuracy rate 1% would yield an increase in annual
earnings of at least N54 million.

1.3 AIMS AND OBJECTIVES

4
  To carry out a systematic research of how prediction accuracy is affected by the
frequency at which the RFM values are calculated with a given transaction data
 To add other user parameters e.g. (age, location) for feature extraction and further
development of recommender system
 To develop a customer profitability model using SVM
 To evaluate the performance of developed customer profitability model.

1.4 SCOPE OF THE STUDY

The scope of this research work is to lay a firm foundation to develop a comprehensive feature

data mining process that starts from definition, extraction and the study of impact these features

have on customer churn and profitability. This research is limited to prediction of customer churn

and profitability using feature data mining which Artificial neutral network and Support vector

machine.

1.5 SIGNIFICANCE OF THE STUDY

These research study capture a complete list of implicit and explicit customer footprints through

feature matrix that enables better prediction of customer churn and profitability. We finally

intend to come up with an end to-end framework and make it available for the organization and

the academic community for predicting customer churn and profitability for e-commerce

business model.

5
 CHAPTER TWO

LITERATURE REVIEW

2.0 INTRODUCTION

Rajini & Sangamaheswary argued that telecom companies must enhance their trust worthiness by

holding customers at heart, offering customized services and listening to their customers to make

every interaction of their customers a lifelong experience. Aggressive customer centric strategies

can only help the telecom companies to survive and retain their market share. In the highly

dynamic business environment, companies with customer focused strategies can only win the

battle for customers. Knowledge about the choices and preferences of the customers help telecom

companies to improve quality of services, better network coverage, relationship development,

price perception, brand image, trust and customer expectations. This not only helps in

maximizing customer satisfaction but also leads to increased loyalty (Sangamaheswary, 2016).

2.1 CUSTOMERS RELATIONSHIP MANAGEMENT (CRM)

CRM can be defined as a set of business activities to increase business performance in

customer management. Customer demands can be changed with time variation. CRM

relationship is understandable by customer life cycle or customer lifetime. The goal of CRM is

to ensure customer satisfaction and delight at every level of interface with the company.‘CRM’

refers to managing relationship with customers. It is a process or method used to learn more

about customers need and behaviors in order to develop strong relationships with customers. It

is type of Management, which is used to satisfy the customer needs. The loyalty of customers

depends on their satisfaction of product or service.

6
CRM helps to manage the churn customers in the company. It helps to attract the new customers.

There exist four dimensions of CRM - Customer Identification, Customer attraction, Customer

Retention and Customer development (K.Rodpysh and M. Majdi, 2012).

• Customer identification: CRM begins with acquiring customers to the company by

identifying them. This is phase where people want to become customers or most profitable

persons for the company. This is basically related to group of customers as they may lead to

profit or loss of company.

• Customer attraction: This phase of CRM helps to make long relationship with their

customer by providing them numerous offers such as discount on product, free products etc.

Customer attraction depends on the satisfaction of products. Satisfied persons help to

increase the retention rate by providing positive information to new employees. It becomes

major role to attract the new customers as churn customers likely to move from one company

to another.

• Customer retention: It occurs when company fulfill the needs of customers. A customer

can retain themselves in a company only when their needs are fulfilled or they are satisfied

with the service given by existing company. To retain the customers in the company, it

becomes necessary to complete the demands of the customers.

7
 Customer Customer
identificatio retention
n

CRM
Cross
Selling
Up-
Customer selling
Customer
attraction developm Value
ent
Analysis

Fig2 : CRM Framework

CRM process can be defined as activities to manage the customer relations by grouping them

according to behavior. It helps to make perceptions by identifying customers, customer

knowledge preparation and by building customer relationships. Basically CRM process can be

grouped into different parts- Vertical & Horizontal process, Front & Back office and Primary &

Secondary process and three levels- Customer Facing level, Functional level and Companywide

level into three levels- Customer Facing level, Functional level and Companywide level.

2.2 MACHINE LEARNING TECHNIQUES-CLASSIFICATION

In the following, we briefly present five well established and popular techniques used for

churn/profitability prediction, taking into consideration reliability, efficiency and popularity in

the research community.

2.2.1. ARTIFICIAL NEURAL NETWORK

The churn / profitability prediction and management is found to be a major issue and the author

(Tsai, 2015) has developed a hybrid model comprising the combination of ANN and ANN

8
(Artificial Neural Network) for pre-processing and combination of Self Organised Maps (SOM)

and ANN for predictive analysis. The single layer and multilayer perceptron is found to be

effective during the training set and the problems of clustering can be easily analysed through

this process. The author has considered two stage process of ANN to achieve higher rate of

accuracy during the training phase and SOM and ANN for analysing clustering and to achieve

higher prediction accuracy. The result obtained from the study shows that an accuracy rate of

93.06% is achieved for testing model and 94.32% during training. In general, it is observed that

the developed model successfully predicts the type-1 and type-2 error and better results is

observed at the training stage rather than testing stage. Furthermore, the combination of K-means

technique for clustering and Multi-Layer Perceptron based Artificial Neural Network (MLP-

ANN) for prediction is developed by author (Hudaib, 2015). The author has developed three

models namely k-means, hierarchical and SOM on the basis of real time churn data and

processed through ANN for classification purpose. The result obtained from the study shows that

K-means along with MLP based ANN predicts better accuracy rate of 97.2% with churn rate of

73%. Moreover, the slightly higher churn rate of 94.6%is observed by deploying combination of

SOM and MLPANN model, but the accuracy rate is decreased to 95.9%. Further, the author

(Khodabandehlou, 2017) has considered the analytical behaviour of the customer to predict the

rate of churn using machine learning techniques. Due to the advancement and modification in the

approaches of selling and supply surplus has created a huge amount of competition in the market.

Six stages namely retrieval of customer data and formation of dataset, pre-processing and

selection of key characters, differentiating into training and testing database, usage of effective

machine leaning techniques for prediction model development, comparative analysis followed by

deploying appropriate strategies for performance evaluation is used in the analysis. The result

9
obtained from the study shows that ANN-MLP model provides highest accuracy rate of 97.07%

and F_measure rate of 97.92% compared to ANN-RBF and SVM-RBF.

2.2.2. SUPPORT VECTOR MACHINES

The early warnings of customer churn along with fraud detection and developing an effective

model is found to reduce the financial losses of approximately millions of in several sectors [16].

The advantages of SVM over other techniques are minimal input parameters for optimization,

linearity constrained quadratic problem for obtaining training dataset and development of

technique is mainly dependent on structural risk minimization (Kim, 2005). The under sampling

approach using one class Support Vector Machine (SVM) is discussed to ascertain the churn

prediction in credit card and automobile insurance sector (Sundarkumar, 2015). The study shows

that two set of data with a ratio of 80.0% and 20.0% is extracted which comprises with a details

of 738 fraudulent in insurance sector. Further, the data is classified and processed through one

class SVM for pre-processing and classification. From the study, it is observed that an accuracy

rate of 60.61% is achieved with higher Area Under the Curve (AUC) value of 8728.5.

Furthermore, the author compared the developed technique with the model proposed by author

(Sundarkumar, 2015) and the study on comparative analysis shows that developed under

sampling methodology increases the rate of AUC and reduces the complexity rate compared to

existing techniques. Furthermore, the application of SVM is used in the field of B2B e-

commerce industry to predict the rate of churn and through unique testing and forecasting

models (Gordini, 2019). The advancement in the e-commerce sector has provided a distinct

solution for sharing information between the clients and the business. The author has captured

input dataset from the online marketing database website (Coussement, 2008). The capability of

RBF kernel technique to handle relationship amongst the dependent and independent variable set

10
is used to optimize the layer of hyper plane and the data processed is forwarded through SVM

model for classification. MATLAB software is used for analysis and the result obtained from the

study shows that SVMauc has highest prediction rate of 89.98% with AUC of 88.61, which is

better compared to other existing techniques. From the summarization section, it is observed that

the predictive performance can be further improvised by comprising the staying power of the

system. Furthermore, a predictive model for customer churn by using the selective item sets such

as bill, records of the data and terminals and their features sets is developed by author (Dong,

2017). From the survey conducted by author, it is shown that SVM is found to be better

applicable to analyse customer churn analysis and results obtained from the study shows a

reasonable accuracy rate of 70.6% is obtained for 403280 input data.

2.2.3. DECISION TREES LEARNING

Decision Trees (DTs) are tree-shaped structures representing sets of decisions capable to

generate classification rules for a specific dataset (Lee, 2001), or as Berry and Linoff noted ‘‘a

structure that can be used to divide up a large collection of records into successively smaller sets

of records by applying a sequence of simple decision rules’’ (Clemente, 2010). More descriptive

names for such tree models are Classification Trees or Regression Trees. In these tree structures,

leaves represent class labels and branches represent conjunctions of features that lead to those

class labels. DT have no great performance on capturing complex and nonlinear relationships

between the attributes. Yet, in the customers churn problem, the accuracy of a DT can be high,

depending on the form of the data (Tsiptsis, 2011).

11
2.2.4. NAÏVE BAYES

A Bayes classifier is a simple probabilistic classifier based on applying Bayes’ theorem with

strong (naïve) independence assumptions. A more descriptive term for the underlying probability

model would be independent feature model. In simple terms, a Naïve Bayes (NB) classifier

assumes that the presence (or absence) of a particular feature of a class (i.e., customer churn) is

unrelated to the presence (or absence) of any other feature. The NB classifier achieved good

results on the churn prediction problem for the wireless telecommunications industry (Nath,

2003) and it can also achieve improved prediction rates compared to other widely used

algorithms, such as DT-C4.5 (Kirui, 2014).

2.2.5. REGRESSION ANALYSIS-LOGISTIC REGRESSION ANALYSIS

Regression analysis is a statistical process for estimating the relationships among variables. It

includes many techniques for modeling and analyzing several variables, when the focus is on the

relationship between a dependent variable and one or more independent variables. In terms of

customer churning, regression analysis is not widely used, and that is because linear regression

models are useful for the prediction of continuous values. On the other hand, Logistic Regression

or Logit Regression analysis (LR) is a type of probabilistic statistical classification model. It is

also used to produce a binary prediction of a categorical variable (e.g., customer churn) which

depends on one or more predictor variables (e.g., customers’ features). In the churn prediction

problem, LR is usually used after proper data transformation is applied on initial data, with quite

good performance (Gürsoy, 2010) and sometimes performs as well as DT (Radosavljevik, 2010).

2.3 RELATED WORK

12
In (Zhen-Yu, 2012) authors study, a framework with ensemble techniques is presented for

customer churn prediction

directly using longitudinal behavioral data. A novel approach called the hierarchical multiple

kernel support vector machine (HMK-SVM) is formulated. A three phase training algorithm for

the H-MK-SVM is developed, implemented and tested. H-MK-SVM constructs a classification

function by estimating the coefficients of static and longitudinal behavioral variables in the

training process without transformation of the longitudinal behavioral data. The training process

of the H-MK-SVM is a feature selection and time subsequence selection process because the

sparse nonzero coefficients correspond to the variables selected.

In (Chih-Fong Tsai, 2009) author considered the two hybrid models by combining two different

neural network techniques for churn prediction, which are back-propagation artificial neural

networks (ANN) and self-organizing maps (SOM). The two hybrid models are, ANN combined

with ANN (ANN + ANN) and SOM combined with ANN (SOM + ANN). In particular, the first

technique of the two hybrid models performs the data reduction task by filtering out

unrepresentative training data. The outputs as representative data are then used to create the

prediction model based on the second technique. To evaluate the performance of these models,

three different kinds of testing sets are considered.

In (Mao-Yuan Chen,2010) authors presented an important process of developing MOD

customer churn prediction models by data mining techniques. The process consists of

preprocessing stage for selecting important variables by association rules that have not been

applied before the model construction stage by neural networks (NN) and decision trees (DT)

13
and four evaluation measures including accuracy, precision, recall, and F-measure all of which

have not been used to examine the model performance. The source data is based on one

telecommunication company providing MOD services in Taiwan, and the experimental results

showed that using association rules allows the DT and NN models to provide better prediction

performances over a chosen validation dataset. In particular, the DT model performs well when

compared with NN model. Moreover, some useful and important rules in the DT model showed

the various factors affecting a high proportion of customer churn.

In (Huang, 2010) authors proposed a new set of features with three new input window

techniques. The new features are demographic profiles, account information, grant information,

Henley segmentation aggregated call details, line information, service orders and bill and

payment history. The basic idea of the three input window techniques is to make the position

order of some monthly aggregated call-detail features from the previous months in the combined

feature set for testing as well as for training phase. For evaluating these new features and

window techniques, the two most common modeling techniques namely decision trees and

multilayer perception neural networks and one of the most promising approach like support

vector machines are selected as predictors.

In (Wouter Verbeke, 2011) authors presented a customer churn prediction models to detect

customers with a high propensity to attrite. Prediction accuracy, comprehensibility and

justifiability are three key aspects of a churn prediction model. An accurate model allow to

correctly targets future churners in a retention marketing campaign, while a comprehensible

and intuitive rule-set allows identifying the main drivers for customers to churn and to develop

14
an effective retention strategy in accordance with domain knowledge. The authors provided an

extended overview of the literature on the use of data mining in customer churn prediction

modeling which showed that only limited attention has been paid to the comprehensibility and

the intuitiveness of churn prediction models. Hence, two novel data mining techniques are

applied to churn prediction modeling and benchmarked to traditional rule induction techniques

such as C4.5 and RIPPER. Both Ant-Miner+ and ALBA induced an accurate as well as

comprehensible classification rule-sets. Ant-Miner+ is a high performing data mining

technique based on the principles of Ant Colony Optimization that allows to include domain

knowledge by imposing monotonicity constraints on the final rule-set.

Ning Lu (2014) designed the use of boosting algorithms to enhance a customer churn prediction

model in which customers are separated into two clusters based on the weight assigned by the

boosting algorithm. As a result, a high risky customer cluster has been found. Logistic regression

is used as a basis learner, and a churn prediction model is built on each cluster, respectively. The

experimental results showed that boosting algorithm provides a good separation of churn data

when compared with a single logistic regression model.

Lee (2011) focused on building an accurate and succinct predictive model with the purpose of

churn prediction by using a Partial Least Squares (PLS) based method on highly correlated data

sets among variables. They not only present a prediction model to accurately predict customers

churning behaviour, but also a simple but implementable churn marketing program was

employed. The proposed methodology allows the marketing managers to maintain an optimal

(atleast a near optimal) level of churners effectively and efficiently through the marketing

programs. Here, PLS is employed as the prediction modelling method.

15
Bose et al designed a two-stage hybrid models to combine unsupervised learning technique with

supervised learning technique. It developed a model for the prediction of customer churn. The

important decision is the separation of churners from non-churners in customer churn

management. Decision tree model are very popular in prediction of churn. It used multiple

variables for clustering and examines different hybrid approaches for utilizing the results of

clustering in order to build supervised learning models for prediction of churn. In the hybrid

method, clustering used as a first stage and decision tree used as a second stage. C5.0 decision

tree models with boosting improved the performance of models in term of top decides lift (Chen,

2009).

Wouter and Verbeke designed the application of Ant-Miner+ and ALBA algorithms on a

publicly available churn prediction dataset in order to build accurate as well as comprehensible

classification rule-sets churn prediction models (Wouter, 2011). AntMiner+ is a high performing

data mining method based on the principles of Ant Colony Optimization which allows to include

domain knowledge by imposing monotonicity constraints on the final rule-set. The advantages of

Ant-Miner+ are high accuracy, comprehensibility of the generated models and the possibility to

demand intuitive predictive models. Active Learning Based Approach (ALBA) for SVM rule

extraction is a rule extraction algorithm, which combines the high predictive accuracy of a non-

linear support vector machine model with the comprehensibility of the rule set format.

Dr. M. Balasubramanian and M. Selvarani (2014) used KDD (Knowledge Discovery in Data

Mining) to hold the churn members in the company, as competitor’s increases at high rate. KDD

is performed with using two categories hypothesis and Discovery Oriented. The limitation arise

with the model is that it only helps to predict churn customers, not to determine the appropriate

16
solution to retain them in company. The methodologies used by them were Data Acquisition,

Data Preparation, Derived Variable and Variable extraction. With the selection of appropriate

attributes and fixing threshold values, accurate results may be produced.

M. Lapczynski (2014) developed a hybrid model of C& RT logit model by integration of

decision tree and logistic model.Hybrid model produce improved results than basic logistic

model as it used decision tree with it. It provider better results when compared to single DT. The

hybrid approach also helps to obtain different probabilities of each test case. Decision trees help

to detect lack of data and logistic regression extend the interpretation. At each test, unique

predicted probabilities are obtained. The limitation that arises during research is that Hybrid C&

RT -logit can be applicable only to Single Decision tree.

S.Sonia and Dr. C. Nalini( 2014) used Mapreduce to predict the churn customers in telecom

industry, Mapreduce and HDFS( Hadoop Distributed File system) helped to mine the large

dataset. The use of hadoop MapReduce resolved the problems of data mining. MapReduce

used to provide the good performance in the form of reliability; scalabity and efficiency.

Mapreduce helps to reduce data size, hence in reducing complexity. NameNode, and the

DataNodes for

HDFS JobTracker and the TaskTracker nodes For MapReduce nodes are used for analysis.

Hadoop NameNode is converted into Hadoop Distributed File System. HMapReduce help to

predict churn customers that lead to produce customized approach for retention methods.

Manjari Anand et al. (2014) implemented ART (Artificial Resonance Theory) algorithm to

perform the customer classification based on the choices. The dataset was taken from the

17
company having vehicles on sale. Classification of customers can be implemented using ART

algorithm and was compared with back propagation algorithm. ART algorithm was proved to

as better algorithm for the classification customers and found to use less time for customer

classification. This algorithm used less time complexity than propagation algorithm. ART

algorithm provides the best time complexity and was implemented in MATLAB.

Manjeet kaur and Dr.Kawaljeet Singh (2013) elaborated guidepost on exchanging unnecessary

client information of a bank into effective and useful information with DM techniques like

naive bayes, decision trees and SVM to pick out important client features to predict churn .

The methodology is made of data sampling, data preprocessing, model construction, and model

evaluation phases . Churn rate success is greater than loyal class when prediction is made. The

prediction of loyal customers is more than churn customers when all algorithms are analyzed.

N.Kamalraj and Dr.A.Malathi(2013) determine the possible churners using the predictive data

mining model. The main goal of the research is to get the complete investigation about the data

analysis in the critical process to precede the successful data mining application. It is used to

investigate the data analysis; robust predictive model can be built by discovering the significant

churn factors. It also examines in keeping the predictive models to make the mobile operators

in order to perform them accurately. The techniques are used for the large data sets of the

telecommunications industry.

R .Obiedat and O. Harfoushi (2013) implemented Hybrid approach of K-mean clustering and

Genetic Programming to predict churn customers. K-mean clustering is used to filter the

dataset and Genetic Programming helps to classify the customers into churners and non-

churners. Four clusters are to be used out of two clusters are discarded. Selected classifiers are

18
loaded into model and results are compared with C4.5, ANN and GP with accuracy and churn

rate. The accuracy rate does not classify exact churn and loyal customers, which is main

limitation of this hybrid approach.

S.Nabavi and S.Jafari(2013) implemented CRISP-DM (

Cross Industry Standard Process for Data Mining) on RFM ( Recency, Frequency and Monetary)

with two different technologies named RandomForest and Boosted technique on the dataset of

Solico food industry. The churn customers are to be predicted and effective measures are to be

planned in retaining them. It helps to identify churn customers using customer behavior analysis

and variables. Among all variables, LOR, Relative Frequency and average inter purchase time

are best predictors for churn prediction.

Dr.U. Devi and S.Madhavi (2012) used CART (Classification and Regression Tree) and C5.0 on

the Dataset of Bank customers. Trees are grown and then pruned back. With CART, it becomes

easy to split data into binary and make patterns for remaining data. Data Mining is used to

convert raw Data into useful information. CART provides high success rate of churn class and

C5.0 gives high success rate of active class. Effective churn prediction model help to attain

benefits from its efforts. Holding 5 percent of old customers can increase the profit of company

by 25 percent.

19
REFFERENCE

V. N. Vapnik, The Nature of Statistical Learning Theory, Springer-Verlag,

NewYork Inc., NY, USA, 1995.

I. Guyon, J. Weston, S. Barnhill, V.N. Vapnik, Gene selection for cancer clas-
sification using support vector machines, Machine Learning 46 (1–3) (2002)389–
422.

K.J. Kim, Financial time series forecasting using support vector machines, Neu-
rocomputing 55 (1/2) (2003) 307–319.

S. Ben-David, M. Lindenbaum, Learning distributions by their density levels:a

paradigm for learning without a teacher, Journal of Computer and SystemSciences
55 (1997) 171– 182.

C.-H. Wu, Y. Ken, T. Huang, Patent classification system using a new hybridgenetic
algorithm support vector machine, Applied Soft Computing 10 (4)(2010) 1164–
1177.

S. Chowdhury, J.K. Sing, D.K. Basu, M.

Nasipuri, Face recognition by general-ized twodimensional FLD method and multi-
class support vector machines,Applied Soft Computing 11 (7) (2011) 4282–4292.

H. Azamathulla, Md. Fu-Chun Wu, Support vector machine approach to

forlongitudinal dispersion coefficients in streams, Applied Soft Computing 11 (2)
(2011) 2902–2905.

V.N. Vapnik, Statistical Learning Theory, John Wiley & Sons, New York,
USA,1998.
Zhen-Yu Chen, Zhi-Ping Fan, Minghe Sun, "A hierarchical multiple kernel support
vector machine for customer churn prediction using longitudinal behavioral data,"
European Journal of Operational Research 223 (2012) 461–472.

Chih-Fong Tsai, Yu-Hsin Lu, "Customer churn prediction by hybrid neural networks,"
Expert Systems with Applications 36 (2009) 12547–12553

20
 Chih-Fong Tsai, Mao-Yuan Chen, "Variable selection by association rules for customer
churn prediction of multimedia on demand," Expert Systems with Applications 37
(2010) 2006–2015

B.Q. Huang, T.-M. Kechadi, B. Buckley, G. Kiernan, E. Keogh, T. Rashid, "A new
feature set with new window techniques for customer churn prediction in land-line
telecommunications" Expert Systems with Applications 37 (2010) 365.7–3665

Wouter Verbeke, David Martens, Christophe Mues, Bart Baesens, "Building

comprehensible customer churn prediction models with advanced rule induction
techniques" Expert Systems with Applications 38 (2011) 2354–2364.

P.C. Pendharkar, "Genetic algorithm based neural network approaches for predicting
churn in cellular wireless network services", Expert System Application 36 (2009) 6714–
6720.

Yaya Xie, Xiu Li, E.W.T. Ngai, Weiyun Ying, "Customer churn prediction using
improved balanced random forests", Expert Systems with Applications 36 (2009) 5445–
5449.

Srini S Srinivasan, Rolph Anderson, and Kishore Ponnavolu. Customer loyalty in

ecommerce: an exploration of its antecedents and consequences. Journal of retailing,
78(1):41–50, 2002.

David Gefen. E-commerce: the role of familiarity and trust. Omega, 28(6):725–737, 2000.

Rajini, G., &Sangamaheswary, D.V. (2016). An Emphasize of Customer Relationship

Management Analytics in Telecom Industry. Indian Journal of Science and Technology,
2016, 9(32), 1-5.

Lu, N., Lin, H., Lu, J., & Zhang, G. (2014). A customer churn prediction model in telecom
industry using boosting. IEEE Transactions on Industrial Informatics, 2014, 10(2), 1659-
1665.

21
Lee, H., Lee, Y., Cho, H., Im, K., & Kim, Y. S. (2011). Mining churning behaviors and
developing retention strategies based on a partial least squares (PLS) model. Decision
Support Systems, 2011, 52(1), 207-216.

Bose, I., & Chen, X. (2009). Hybrid models using unsupervised clustering for prediction of
customer churn. Journal of Organizational Computing and Electronic Commerce, 2009,
19(2), 133-151.

Verbeke, W., Martens, D., Mues, C., &Baesens, B. (2011). Building comprehensible
customer churn prediction models with advanced rule induction techniques. Expert Systems
with Applications, 2011, 38(3), 2354-2364.

Sundarkumar, G. G., Ravi, V., &Siddeshwar, V. One-class support vector machine based
undersampling: Application to churn prediction and insurance fraud detection. In
Computational Intelligence and Computing Research (ICCIC), 2015 IEEE International
Conference on(pp. 1-7). IEEE (2015).

Sundarkumar, G. G., & Ravi, V. A novel hybrid undersampling method for mining
unbalanced datasets in banking and insurance. Engineering Applications of Artificial
Intelligence, 37, 368-377 (2015).

Gordini, N., &Veglio, V. Customers churn prediction and marketing retention strategies. An
application of support vector machines based on the AUC parameter-selection technique in
B2B e-commerce industry. Industrial Marketing Management, 62, 100-107 (2017).

Coussement, K., & Van den Poel, D. Churn prediction in subscription services: An
application of support vector machines while comparing two parameter-selection techniques.
Expert systems with applications, 34(1), 313-327 (2008).

22
Kim, S., Shin, K. S., & Park, K. An application of support vector machines for customer
churn analysis: Credit card case. In International Conference on Natural Computation (pp.
636-647). Springer, Berlin, Heidelberg (2005).

Dong, R., Su, F., Yang, S., Cheng, X., & Chen, W. Customer Churn Analysis for Telecom
Operators Based on SVM. In International Conference on Signal And Information
Processing, Networking And Computers (pp. 327-333). Springer, Singapore (2017).

Tsai, C. F., & Lu, Y. H. Customer churn prediction by hybrid neural networks. Expert
Systems with Applications, 36(10), 12547-12553 (2009).
Hudaib, A., Dannoun, R., Harfoushi, O., Obiedat, R., &Faris, H. Hybrid data mining models
for predicting customer churn. International Journal of Communications, Network and
System Sciences, 8(05), 91 (2015).

Khodabandehlou, S., and Zivari Rahman, M. Comparison of supervised machine learning

techniques for customer churn prediction based on analysis of customer behavior. Journal of
Systems and Information Technology, 19(1/2), 65-93 (2017).

C. Kirui, L. Hong, W. Cheruiyot, H. Kirui, Predicting customer churn in mobile telephony

industry using probabilistic classifiers in data mining, Int. J. Comput. Sci. Iss. (IJCSI) 10 (2)
(2013) 165–172.

G. Kraljevic´, S. Gotovac, Modeling data mining applications for prediction of prepaid churn
in telecommunication services, AUTOMATIKA: cˇasopis za automatiku, mjerenje,
elektroniku, racˇunarstvo i komunikacije 51 (3) (2010) 275–283.

R.J. Jadhav, U.T. Pawar, Churn prediction in telecommunication using data mining
technology, IJACSA Edit. 2 (2) (2011) 17–19.

23
D. Radosavljevik, P. van der Putten, K.K. Larsen, The impact of experimental setup in
prepaid churn prediction for mobile telecommunications: what to predict, for whom and does
the customer experience matter?, Trans MLDM 3 (2) (2010) 80–99.

Y. Richter, E. Yom-Tov, N. Slonim, Predicting customer churn in mobile networks through

analysis of social groups, SDM, vol. 2010, SIAM, 2010, pp. 732–741.

S. Gürsoy, U. Tug˘ba, Customer churn analysis in telecommunication sector, J. School Bus.

Admin. Istanbul Univ. 39 (1) (2010) 35–49.

G. Klepac, Developing Churn Models Using Data Mining Techniques and Social Network
Analysis, IGI Global, 2014.

D.T. Larose, Discovering Knowledge in Data: An Introduction to Data Mining, John Wiley &
Sons, 2014.

A. Sharma, D.P.K. Panigrahi, A neural network based approach for predicting customer
churn in cellular network services, Int. J. Comput. Appl. 27 (11) (2011) 26–31.
Learning Theory, ACM, 1992, pp. 144–152.

S.J. Lee, K. Siau, A review of data mining techniques, Ind. Manage. Data Syst. 101 (1)
(2001) 41–46.

G.S. Linoff, M.J. Berry, Data Mining Techniques: For Marketing, Sales, and Customer
Relationship Management, John Wiley & Sons, 2011.

24