Telecommunication Industry Customer Churn Computer Science Essay

Published:

Customer churn is a business term used to describe the loss of customer. It describes those customers or clients who leave or switch to competitors. In the telecommunication industry, customers have multiple choices of services and they frequently switch from one service to another. In this competitive market, customers demand best products and services at low prices, while service providers constantly focus on getting hold of as their business goals. So that's why there is very higher rate of customer churn in telecommunications industry experiences an average of 30-35% annual churn rate. The purpose of this paper is to propose an efficient Customer Churn Prediction Model based on classification techniques, which will help the telecommunication company to predict the customer churn rate to know about which customers are loyal to them.

INTRODUCTION

Data mining is very famous technique for churn prediction and it is used in many fields. Data mining refers to the process of analyzing data in order to determine patterns and their relationships. It is an advanced technique which goes deep into data and uses machine learning algorithms to automatically shift through each record and variable to uncover the patterns and information that may have been hidden. There is a lot of work done in data mining for churn prediction in different fields. It is used to solve the customer churn problem by identifying the customer behavior from large number of customer data.

Problem Statement:

Lady using a tablet
Lady using a tablet

Professional

Essay Writers

Lady Using Tablet

Get your grade
or your money back

using our Essay Writing Service!

Essay Writing Service

Customer churn refers to the periodic loss of customers in an organization. Customer churn is a very common problem of every organization all over the world. In the competitive market it's a very big challenge for any organization to retain their customers and build a valuable place in the market to gain more customers. In Telecommunication Companies it is very lavish process because users are switching frequently from one service to another according to their interests and needs and market competition growing increasingly. Pakistani organizations are also facing this problem. So we are going to build a classification model for churn prediction in telecommunication companies of Pakistan. This model will work on local data of telecom customers of Pakistan.

The paper will provide an application to telecommunication companies who want to know about their customers and details of their customers that who are existing and loyal to them and who are going to leave or quitting from their products. Also to know about who are loyal to them so that they can give them some extra benefits to be their loyal customers and make better campaign for customer retention.

There are different classifications models designed for different organization which are predicting and recommending those organizations to save their customers/employees by defining new policies. They use different data mining classification modeling techniques like SVM, decision trees, Naïve Bayes algorithm, mining mart, life time values techniques which are producing different results and generating patterns in different manners. So we are going to develop an efficient classification model to predict the customer churn problem using Neural Network and fuzzy classification techniques [1].

In section 1 we have given a short introduction of data mining and customer churn problem. The remaining structure of our research is as follows.

In section 2 there is a brief description of background study. Section 3 describes our proposed model for prediction using neuro-fuzzy technique. Section 4 has validation and discussions and finally in section 5 there is a brief conclusion of our research and future work description.

background study

Data Mining:

Data mining is a process that uses various techniques to discover the hidden patterns or knowledge from the pool of data. It is a powerful tool to transform the data into information. It predicts future trends and behaviors [2].

It is widely using in many business problems like identifying customer behavior, fraud detection, finding natural segments, predicting which customer are likely to churn etc using different classification techniques. The most important techniques used for data mining are neural networks, decision trees, fuzzy logic, nearest neighbor, SVM, genetic algorithms, rule induction etc. Data mining applications are widely used in many fields of scientific research, businesses, banking sector, intelligence agencies etc [3].

Classification in Data mining:

Lady using a tablet
Lady using a tablet

Comprehensive

Writing Services

Lady Using Tablet

Plagiarism-free
Always on Time

Marked to Standard

Order Now

Classification is a data mining process which arranges the data into predefined groups. There are different classification algorithms include logistic regression, robust regression, decision tree, nearest neighbor, naive Bayesian classification and neural networks [3].

Neural network (NN) is a classification technique used in data mining that resembles the biological neurons and is a non linear predictive model that learns through data training for prediction. In data mining there are different neural network architectures used for mining like multilayered feedfarward neural networks, back propagation neural networks, Kohonen's self organizing maps [4].

Fuzzy logic is also a very good technique used in data mining. The fuzzy approach enables approximate reasoning and it is suitable for modeling human decision process [4].

Customer Churn Problem:

Customer churn is a very common in business community. Churning is a very costly process in an organization because it is more difficult to gain new customers than retaining existing one [1].

Table 1: Show the examples of churn prediction in literature.

Article

Market sector

Case data

Method used

Morik, K., Scholz, M. [1]

Insurance company

163,745 customers

TF/IDF

Au et al. [2]

Wireless telecom

100 000

subscribers

DMEL method (DM using Evolutionary Techniques )

Buckinx et al. [4]

Retail business

158 884

customers

Logistic regression, ARD and decision tree

Buckinx et

al. [5]

Daily grocery

878 usable

responses

MLR (regression), ARD, and

decision tree

Ferreira et

al. [10]

Wireless telecom

100 000

subscribers

Neural network, decision tree, hierarchical Neurofuzzy systems, rule evolver

Gatland [11]

Retail banking

1 100 customers

Multiple regression

Hwang et al. [14]

Wireless

Telecom

16 384

customers

Logistic regression, neural network,

decision tree

Mozer et al.

[16]

Wireless

telecom

46 744

subscribers

Logistic regression, neural network,

decision tree

Artificial Neural Networks (ANN):

ANN is a mathematical computational model that tries to simulate the structure of biological neural networks. It composed of interconnected group of artificial neurons using computational approach [5]. The basic idea behind artificial neural networks is that each attribute is associated with a weight and combinations of weighted attributes participate in the prediction task. During learning the weights are constantly updated, thus correcting the effect which an attribute has. Given a customer data set and the set of predictor variables the neural network tries to calculate a combination of the inputs and to output the probability that the customer is a churner. Back Propagation Neural Network is an artificial neural network technique [6].

The diagram shows Back Propagation neural networks:

Figure 1: Basic structure of NN [5]

This neural network consists of three layers:

Input layer ( 3 Neurons)

Hidden layer (2 Neurons)

Output layer (2 Neurons).

The output of a neuron in a layer goes to all neurons in the following layers. A particular neuron holds input weights (weights assumed to be 1). Therefore the input values can not be changed. The Back Propagation NN should have two layers input as well as output layers but may contain zero or more hidden layers [6].

The back propagation algorithm is a supervised learning algorithm, which perform minimization of error by gradient descent technique. A data is trained with back propagation network. When a target output pattern exists, the actual output pattern is computed. The gradient descent acts to adjust each weight in the layers to reduce the error between the target and actual output patterns. The adjustment of the weights is collected for all patterns and finally the weights are updated [7].

Neuro fuzzy:

Fuzzy logic and neural networks are natural complementary tools in building intelligent systems. Integrated neuro-fuzzy systems can combine the parallel computation and learning abilities of neural networks with the human-like knowledge representation and explanation abilities of fuzzy systems. As a result, neural networks become more transparent, while fuzzy systems become capable of learning [7].

Neuro-fuzzy is the combinations of artificial neural networks and fuzzy logic. Neuro-fuzzy system incorporates the human-like reasoning style of fuzzy systems through the use of fuzzy sets and a linguistic model consisting of a set of IF-THEN fuzzy rules and with the learning and connectionist structure of neural networks [8].

Lady using a tablet
Lady using a tablet

This Essay is

a Student's Work

Lady Using Tablet

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

Examples of our work

Figure 2: Basic structure of Neuro Fuzzy Expert system [7]

proposed model

Churn Prediction Methodology:

The methodology we are going to follow for churn prediction is data mining methodology. The first step is to understand the business problem. We have to analyze and define the business problem i.e. churn prediction in telecom sector. For this purpose we identify and define all those demographics or attributes of customer which are affecting the churn rate of a company. Then we perform the pre-processing on the customer data for data cleansing and transform it into appropriate form. In second step the data mining classification model is building by using Neuro fuzzy technique. The steps involved in our model are as follows [8].

Identifying both input and output values;

Identifying fuzzy sets for input values;

Identifying fuzzy rules;

Creating and training the neural network;

We will train our classification model on historical data to extract the patterns. On the basis of these patterns, the churn rate will be calculated. The whole process of churn prediction is shown in following diagram.

Figure 3: Conceptual Model for Customer Churn Prediction using Neuro fuzzy logic [8]

Data Source:

The data source for prediction will include three types of customer data variables: customer household demographics; usage data (e.g. Minutes use and Revenue); and company interaction data such as customer calls into the customer service center [9].

The customer household demographics includes customer name, id , age , marital status, gender , occupation, estimated income, no. of children, no. of vehicle, no. of credit cards used, etc [10].

The customer usage data includes the internal data of customer from company's data warehouse. The data is about market channel, plan type, customer segmentation code, ownership of the company's other products, dispute, late fee charge, discount, toll free services, weekly average call counts, percentage change of minutes, share of domestic/international revenue etc [11].

The company interaction data such as customer calls into the customer service center includes general inquiry calls, complaint calls, calls for change the service, calls about the cancellation of any service, general inquiry mails, complaint mails, mails send to company for change the service, etc [12].

Following are the interface forms to collect the input demographics of a telecom customer.

Input data source

Customer demographic data

Company Interaction Data

Customer usage data

The neuro fuzzy layered architecture for customer churn classification model is shown as follow.

Figure 4: Neuro Fuzzy Layered Architecture

In Figure 4, the system has five-layer feed forward network architecture. The Layer-1 represents the input layer which describes the customer demographics. Fuzzification layer (Layer 2). The units in this layer have fuzzy membership functions known as transfer functions. The purpose of this layer is to fuzzify the input values - to translate them into fuzzy sets. Each unit in this layer corresponds to a single fuzzy set that appears in the premise part of a fuzzy rule. Fuzzy rule layer is Layer-3. Each unit in this layer corresponds to a certain rule. The output membership layer is described as Layer 4. Neurons in this layer represent the fuzzy sets used in consequent of fuzzy rules. An output membership neuron combines all its inputs using fuzzy union operation (in the case of customer churn classification one for each type of customer, e.g., churn, highly churn, not-churn, etc). The defuzzification layer is at level 5 (Layer 5). It takes the output fuzzy sets and combines them into a single fuzzy set (i.e. churn or not churn).

A neuro-fuzzy system is essentially a multi-layer neural network, and we have applied standard learning algorithms developed for neural networks, including the back-propagation algorithm. When training input-output customer demographics are presented to the system, the back-propagation algorithm computes the system output and compares it with the desired output of the training data [10]. The error is propagated backwards the neuron activation functions are modified as the error is propagated. The propagation algorithm differentiates the activation functions. Given input and output linguistic values, a neuro-fuzzy system will automatically generate a complete set of fuzzy IF-THEN rules.

The data preprocessing was performed on input data sources and separate all those attributes which have high influence on customer churn. Then we select the linguistic variables from those attributes and find the relationship between linguistic variables and attributes to make the fuzzy sets and defined fuzzy rules.

Following we are presenting few linguistics variables and fuzzy sets for telecom customer to predict the churn rate.

We have defined the fuzzy sets qualifier (known as hedges) Y, Y* and N for linguistic variables of customer data [14].

Y = (0 < µ(x) < 1)

Y* = (µ(x) = 1)

N = (µ(x) = 0)

Here we are finding the relationship between customer age and mobile services that which age group is using which service highly (e.g. SMS, calls, mobile internet service, fun services, seasonal services etc) and customer income that which services satisfied the customer according to his/her income. The degree of membership for fuzzy sets is between (0-1).

The table shows the fuzzy set for customer age and mobile services.

Age

SMS

Calls

Mobile Internet

Fun Services

Seasonal services

15-20

Y*

N

N

Y

Y

20-25

Y*

Y

Y

Y

Y

25-30

Y

Y

Y

N

Y

30-35

Y

Y*

Y

N

N

35-40

Y

Y*

Y

N

N

40-45

Y

Y*

N

N

N

45-50

Y

Y*

N

N

N

50-60

N

Y*

N

N

N

The rules are defined for fuzzy set to find the impact of age on mobile services for prediction the churn rate. For example

If Age is 15-20 & Service is SMS

Then mobile usage is HIGH.

If Age is 35-40 & Service is Calls

Then mobile usage is HIGH.

If Age is 40-45 & Service is Fun

Then mobile usage is LOW.

and so on. We defined different rules like above and see the impact of customer demographic on mobile services.

The following graph shows the usage rate of mobile services by different age group customers.

The following table shows the fuzzy set for customer income and mobile services

Income

SMS

Calls

Mobile Internet

Fun Services

Seasonal services

Below 5000

Y*

Y

N

N

Y

5000-10000

Y*

Y

N

N

Y

10000-20000

Y*

Y

N

N

Y

20000-30000

Y*

Y*

N

Y

Y

30000-40000

Y

Y*

Y

Y

Y

40000-50000

Y

Y*

Y

Y

Y

Above 50000

Y

Y*

Y

Y

N

Same here the rules are defined for fuzzy set to find the impact of income on mobile services for prediction the churn rate. For example

If Income is less than 5000 & Service is SMS

Then service use is HIGH.

If Income is 5000-10000 & Service is Calls

Then service use is AVERAGE.

If Income is above 50,000 & Service is Calls

Then service use is HIGH.

and so on…

The following graph shows the usage rate of mobile services having different ranges of income

validation and discussions

To find the good results of customer churn prediction in telecommunication, we used neuro fuzzy logic and back propagation neural networks techniques. It produced good results by training the system with customer data. We have implemented the model in java programming language using tool java NetBean IDE 6.5.1 and at backend the PostgreSQL 8.4 database.

The churn prediction of customer is analyzed by constructing a classifier. It uses a sufficiently large data set that contains churning and non-churning customers. The work of a classifier is to decide, given a customer data set, if churn is more or less likely.

Actual Churn

Actual Non-Churn

Predicted Churn

True

False

Predicted Non-Churn

False

True

Customer Churn

Customer Non-Churn

0 1

1 0

The results of churn are in shown in table below

Actual churn

Predicted churn

Total

Non-churn

Churn

Non-Churn

385

93

449

Churn

104

117

221

Total

490

210

700

Here is a comparison of different classification techniques used for churn prediction.

Model

Actual

Predicted

Non-churner

churner

Neural Networks

Non- churner

70%

30%

Churner

35%

65%

Fuzzy Logic

Non-churner

69%

31%

Churner

40%

60%

Neuro Fuzzy

Non- churner

58%

42%

Churner

33%

67%

CONCLUSION AND FUTURE work

This paper deals with the customer churn analysis in telecommunication industry of Pakistan. Artificial neural network and fuzzy logic is used separately in solving different problem domains. These are very useful classification techniques for prediction. We have used both combined techniques i.e. Neuro-fuzzy for prediction of customer churn in telecom companies. This technique used to minimize the total error of the network to predict two different types of customers i.e. churn customers and non-churn customers. The model will be used to predict the churn for the telecom customers. Thus the study successfully developed a framework using available database technologies that can help the telecommunication industries to get the idea about their business progress and it would be helpful for making retention strategies.