 Abstract

A trendy topic or indicator in the telecom industry is the consumer churn rate. There is a huge amount of
data generated every minute in most of the industries especially in telecommunication industry. In
addressing such common problem in the field like customer churn, data mining techniques are available.
The reasons for churning and attracting customers need to be known by telecom providers, which can be
realized through the information achieved from telecom data.
In today's dynamic environment, one of the most important issues for network operators is consumer churn
and the launch of new packages. There are a few predictive modeling approaches for churn, and these
models can be mapped via data mining processes to consumers with a certain pattern of habits, profiling
and product packaging. Data mining model via the CRISP-DM strategy helps to minimize the rate of
customer churn and retain customers as well.
Keywords: Churn rate, Customers, CRISP-DM strategy

 Introduction

Over the last two decades considerable data volume is rising at a phenomenal rate due to advancements in
Information Technology. In parallel data mining also records experienced tremendous growth. There are various
new methods and techniques are introduced to process the data in collecting information. The process of extracting
useful information from data can be described as data mining. In different domains various data mining approaches
are successfully applied. Meanwhile telecom industries are facing a difficulty in customer churn rate. Hence
customer churn models are intended to classify clients with a high likelihood of jumping or quitting the service
provider.

Databases on customer churn assisting the businesses to resist or decrease the rate of customer churn by enabling
them to initiate retention and remedial actions. Retaining the old customers is up to the decision of businesses
simply it is a choice. It is foreseeable that drawing new customer costs than retaining the old customers by about five
to six times. Recruiting a new client includes cost of hiring a new staffs, promotional expenses, and discounts. A
loyal customer who has been with a brand for a long time appears to produce higher sales and is less vulnerable to
rival costs. (Kraljevíc & Gotovac, 2010)

Telecom industry offers a service to certain average number of customers in case the accounted customer number
falls below due the device’s design which is perceived to be the company's loss. A bit step of maintaining an
existing client will lead to a large increase in sales and profits. The need for customer retention requires detailed
customer churn rate minimization and data mining models that are both precise and understandable. To prevent the
losses to the telecom company, the models have to identify customers who are about to churn and the cause for
churn. A model should be built to identify the reasons for customers churn and the modifications needed to maintain
customers. The data mining model based on the CRISP-DM method for a special mobile package is discussed in the
article here (Ćamilović, 2008).

Related Works (Provide detailed literature survey related to CRISP-DM process)

CRISP-DM strategy addresses the most useful telecommunications data mining applications for Customer relationship
management (CRM). There are articles explaining the value of Telecommunications industry. Many companies are armed with
numerous CRM instruments and processes like CRISP-DM since they are willing to adapt a customer oriented organizational
culture. Analytical CRM which is known to be CRM software uses Online Analytical Processing (OLAP) and data mining. As
per Data mining and CRM IN in telecommunications paper, there are valuable CRM data mining applications in
telecommunications which is applied by them.

CRISP-DM strategy consists effective CRM phases of data mining project as follows;

1
 1. Defining the problem to be solved.
2. Analyzing and preparing the data need to be accessed.
3. Modeling and presenting the data into a data mining software program.
4. Results to be interpreted.
5. Applying the results by changing an organization’s behavior for competitive advantage.

Predictive customer churn modeling based on data mining techniques is explained in a report in which the use of one of its
predictive model techniques, the Decision Tree Analysis model, is explained and illustrated in detail. To construct a model for
churner prediction the report uses data mining techniques(Almana et al., 2014).

A research that is also focusing on the system of the decision tree, which is a tree-shaped structure that describes choices or data
pattern prediction sets. This is also similar for defining the consistency of interrelated decisions or predicting future patterns in
data and has the potential to classify entities into particular groups based on the characteristics of each entities. It implemented
how to use the classification rule decision tree of data mining in the telecommunications industry in order to evaluate the most
common issue of customer churn rate. A specific training sample set is used to conduct an experiment of customer churn factor
using decision tree (Mo et al., 2007).

The research that is considering on voluntary churning. The collection of DM processes used to extract and validate patterns in
data is the key of the information discovery process from a technological point of view. These processes include several phases,
including data collection, preprocessing of data, transformation of data, data mining, and pattern analysis and evaluation. They
attempt to explore the broadband user character in the Pearl River Delta, using data mining techniques, and construct a "input -
Third International Symposium on Information Processing transformation - output" model based on C5.0, logistics regression,
and neural network.

Finally, customer satisfaction comments are deducted. The regulation of client churn is applied to data mining techniques in the
study. They have defined a method for this non-state customer scheme that uses data mining techniques to create an early-
warning model. The strategy is an opportunity for businesses to improve their existing marketing and CRM. The proper
alignment of the abilities of a company with the mission at hand would realize improved effectiveness (Chen et al., 2011)

In another study, the methodology and technique of a customer chum prediction model is thoroughly explored and predictive
results from the technology and business perspective are examined more closely to help reduce customer leaving. This approach
would help 812 customers understand the telecommunications industry thoroughly, get the more precise consumer model, so the
implementation is highly necessary to improve marketing decision-making and customer service (Kraljevíc & Gotovac, 2010).

Another study aimed at developing and implementing a modeling and forecasting system for the selected European
telecommunications industry for churning customers. The proposed scheme could serve as quantitative support for decision-
makers in customer relationship management (CRM). CRM managers and subordinates have access to the tool to help them
identify customers with a high risk of churning over the next 45 days, identify the key reasons for churning, take appropriate
measures to discourage customers from churning, minimize the churn rate, and retain the customer base and, last but not least,
and explore customer groups with similar characters

As per the Proposal and Implementation of Churn Prediction System for Telecommunications Company article, the main
objective of the research is for the selected European telecommunication industry make a proposal on churn prediction method
and implementing it. Hence the motive of the system is helping the businesses to reduce the customer turnover and create mutual
consent with the customers. T further to the research there are basically three (03) technologies are added in the proposal such as;
statistical programming environment R, MySQL relational database management system and Like Sense visualization tool. The
major advantage of the proposal is all three technologies are open source so that the client company can establish the own system
for business and can predict or overview the turnover rate of relativity low cost customers (Mo et al., 2007)

2
 Case Study: AirMobiTel Explain the methodology you are proposing.

Figure 1:Phases of CRISP-DM process

As AirMobiTel planned to introduce a mobile package to the customers in order to attract them and
minimizing the churn rate of customers. Therefore cross industry standard process for data mining
approach is going to be applied for this case.
Considering the CRISP-DM process, it consists of six main steps/phases to conduct the data mining model.
The steps are Business understanding, data understanding, data preparation, modeling, evaluation and
deployment.
In order to conduct the model using the CRISP-DM process, the approach always doesn’t initialize with
business understanding. Based on the requirement it will start with evaluation or data understanding (Pan,
2010).

Business understanding

3
As executing the task the main goal for me is to propose mobile packages to the customers which will be more
beneficial for them. In order to achieve the goal, the objectives will be created. The objectives shall be:
1. Understanding the areas where the users used to spend a lot.
2. Emphasize the advantage of using the packages
3. Explain the offers and package cost in details

To perform this a set of people containing 5 members will be allocated to understand the current demand of internet
using.
They will research on the above mentioned topic and access the business understanding towards delivering new mobile
packages to the customers to minimize the churn rate.

Data understanding

Data understanding - decide what information is available to meet your business needs. The data understanding will be
gained through business understanding. Data are the available raw materials to constraint the model that we need to
propose. From the set of data we are having, we will build an appropriate solution. The solution shall be the packages
that will be proposed in the end. The data chosen should represent a reasonable amount of data within a specified
period of time.

In case of proposing the new mobile packages, the existing data may not be useful for this particular purpose. In terms
of filter out the necessary data, the data cleansing may be too costly.

Data filtering and data cleansing

This task must be done to ensure that we will be constraining a model using the right set of data.

The data collection will be through surveys from the people at the age group of 12-50. Among them the
areas they spend most of the time will be gathered in percentages along with the devices they are using. All
these details will be included for the study and the proper solution could be suggested through a model.
Understanding the data for proposing new mobile packages, we can,
1. Select the churn customer- They need to be identified and proposing the package as fulfilling their
requirements.

2. Select the online customer – They are mostly related with communication field and internet using. In
order to minimize the churn rate, they should be considered.

Data Preparation

Once the needed data are collected, data analyzing will be done. The data analysis will be done using the analytic
technologies. The technologies impose the requirements on data. The data should be in a right form to implement the
process for developing model for mobile packages. The data preparation shall be executed through these methods:
1. Convert data to a right format
2. Remove missing values
3. Handle outliers
4. Convert data to different types

In order to optimize the database efficiency, simplify the analysis and enable specific queries to be performed
easily, the table should be divided into subsets where necessary.

Modeling

As the initial primary stage of modeling is applying the data mining techniques. Then the output result will be
looked to get. The result shall be a pattern do implement consist with the package offerings. Models are all about

4
 designing our data models or solutions according to our needs. From the set of data received from AirMobiTel,
change them as training set and testing set before applying Neural Network algorithm. The Artificial Neural
network will be choosen instead of other data mining techniques as it will provide more accurate results.

The data mining techniques that will be applied in modeling shall be as classification and association in proposing the model for
AirMobiTel. As AirMobiTel wants to develop a data mining model to propose new mobile packages, the model will use the
filtered and cleaned data as follows

1. The areas where most of the mobile users spend time

2. The platform they are using
3. The devices they are using.
If the model is chosen, various parameters are obtained to boost the performance.

Evaluation

Evaluation is all about believing that the gained results are more reliable. Before the deployment the
created model should be tested. Thereafter the results can be deployed after data mining.
This process is done to ensure that the pattern extracted from the data are true and proper ones, and
also Ensure that the model satisfies the business goal. Make sure there is no flaw in the actual business
context.
Evaluation is carried out on the response time, level of trust, cost, error rate and the effectiveness of the
model in achieving the previously established goals and objectives.

Deployment

The purpose of deployment is to provide the service individuals with the effects of prediction in a
certain manner. We must also pay attention to the model problem from the angle of Service. The
forecast results primarily include: the customer's chum name list; the chum ratio. Implement the gained
data mining results in real. Identifying the output from the input given.
Data mining techniques also may be deployed in the model for mobile packages as like
1. Reduce the tasks of modeling from data analysts
2. The demand might change before the data analysts can adapt

 Conclusion
Over the last three to five years, the capabilities for collecting data have been rapidly increasing in all industries,
especially in the telecommunications industry. The telecommunications market is intensely competitive, and
telecommunications companies understand that their key assets are consumers.

In order to gain competitive advantage they have to Understand customers’ behavior and, interact with clients and
provide them with specialized and versatile facilities according to their requirements. In this article a CRISP-DM
data mining process for proposing new mobile packages was discussed. The tasks to be done in each and every
phase of the process is explained thoroughly.

In order to help minimize the customer leaving and retain them, CRISP-DM and more findings from the technology
and market view are thoroughly evaluated according to the traditional data mining procedure. This approach will
assist AirMobiTel to achieve a more specific consumer model, so the application value is extremely essential to
minimize the churn rate and customer service.

5
 References

Almana, A. M., Aksoy, M. S., & Alzahrani, R. (2014). A Survey On Data Mining Techniques In
Customer Churn Analysis For Telecom Industry. Journal of Engineering Research and
Applications, 4(5), 165–171.
Ćamilović, D. (2008). Data Mining and Crm in Telecommunications. Serbian Journal of
Management, 3(1), 61–72.
Chen, Y. B., Li, B. S., & Ge, X. Q. (2011). Study on predictive model of customer churn of
mobile telecommunication company. Proceedings - 2011 4th International Conference on
Business Intelligence and Financial Engineering, BIFE 2011, 114–117.
https://doi.org/10.1109/BIFE.2011.112
Kraljevíc, G., & Gotovac, S. (2010). Modeling data mining applications for prediction of prepaid
churn in telecommunication services. Automatika, 51(3), 275–283.
https://doi.org/10.1080/00051144.2010.11828381
Mo, Z., Zhao, S., Li, L., & Liu, A. J. (2007). A predictive model of churn in telecommunications
based on data mining. 2007 IEEE International Conference on Control and Automation,
ICCA, 00, 809–813. https://doi.org/10.1109/ICCA.2007.4376469
Pan, D. (2010). On customer churn and early warning model of telecom broadband. Proceedings
- 3rd International Symposium on Information Processing, ISIP 2010, 157–160.
https://doi.org/10.1109/ISIP.2010.100