International Journal of Engineering and Technical Research (IJETR)

ISSN: 2321-0869, Volume-3, Issue-5, May 2015

Churn Prediction in Telecom Industry Using R

Manpreet Kaur, Dr. Prerna Mahajan
services, it is one of the reasons that customer churn is a big
Abstract Telecommunication market is expanding day by problem in the industry nowadays. To prevent this, the
day. Companies are facing a severe loss of revenue due to company should know the reasons for which the customer
increasing competition hence the loss of customers. They are decides to move on to another telecom company. It is very
trying to find the reasons of losing customers by measuring difficult to keep customers intact for long duration as they
customer loyalty to regain the lost customers. The customers
move to the service that suits most of their needs.
leaving the current company and moving to another telecom
company are called churn.
The research paper is using data mining technique and R 1.2 Types
package to predict the results of churn customers on the Telecom Churns can be classified in two main categories:
benchmark Churn dataset available from Involuntary and Voluntary. Of the two, Involuntary are easier
(http://www.dataminingconsultant.com/data/churn.txt). The R to identify. Involuntary churn are those customers whom the
tool has represented the large dataset churn in form of graphs Telecom industry decides to remove as a subscriber. They are
which depicts the outcomes in various unique pattern churned for fraud, non-payment and those who dont use the
visualizations. The Churn Factor is used in many functions to service. On the other hand, Voluntary churn are difficult to
depict the various areas or scenarios where churners can be
determine, here it is the decision of the customer to
distinguished. The paper is considering churn factor in account
to depict various patterns for churners. R is a powerful
unsubscribe from the service provider. Voluntary churn can
statistical programming tool which can represent the dataset further be classified as incidental and deliberate churn. The
graphically with respect to different parameters and it also uses former occurs without any prior planning by the churn but due
different packages available. to change in the financial condition, location, etc. Whereas,
Churns can be reduced by analyzing the past history of the the latter happens for technological advancement, economics,
potential customers systematically. In the past few years, the fast quality factors and convenience reasons. Most operators are
emerging requirements from both academia and industry has trying to deal with these type of churns mainly.
helped R programming language to emerge as one of the
necessary tool for visualization, computational statistics and 1.3 Managing Churns
data science
Churn management is very important for reducing churns as
acquiring a new customer is more expensive than retaining the
Index TermsChurn, R Tool, Telecommunication, Data
mining.
existing ones. Churn rate is the measurement for the number
of customers moving out and in during a specific period of
I. INTRODUCTION time. If the reason for churning is known, the providers can
then improve their services to fulfill the needs of the
Numerous telecom companies are present all over the world. customers.
Telecommunication market is facing a severe loss of revenue Churns can be reduced by analyzing the past history of the
due to increasing competition among them and loss of potential customers systematically. Large amount of
potential customers. Many companies are finding the reasons information is maintained by telecom companies for each of
of losing customers by measuring customer loyalty to regain their customers that keeps on changing rapidly due to
the lost customers. To keep up in the competition and to competitive environment. This information includes the
acquire as many customers, most operators invest a huge details about billing, calls and network data. The huge
amount of revenue to expand their business in the beginning. availability of information arises the scope of using Data
Therefore, it has become important for the operators to earn mining techniques in the telecom database. The information
back the amount they invested along with at least the available can be analyzed in different perspectives to provide
minimum profit within a very short period of time. various ways to the operators to predict and reduce churning.
Only the relevant details are used in analysis which contribute
1.1 Churn Prediction to the study from the information given.
Churn in the terms of telecommunication industry are the Data mining techniques are used for discovering the
customers leaving the current company and moving to another interesting patterns within data. One of the most common data
telecom company. With the increasing number of churns, it mining technique is Classification, its aim is to classify
becomes the operators process to retain the profitable unknown cases based on the set of known examples into one
customers known as churn management. In of the possible classes. Here, in case of telecom churn,
telecommunication industry each company provides the Classification helps learn to predict whether a customer will
customers with huge incentives to lure them to switch to their churn or not based on customers data stored in database.

Ms.Manpreet Kaur, Student, Guru Gobind Singh Indraprastha II. BACKGROUND

University, Institute Of Information Technology & Management, MCA,
Trainee at NIIT TECHNOLOGIES, DELHI , INDIA. 2.1. Data Mining Techniques
Dr. Prerna Mahajan, Professor, Department of Computer The process of reducing, analyzing the patterns, predicting
Science,Guru Gobind Singh Indraprastha University/ Institute of the hidden and useful required information from large
Information Technology & Management, Delhi/India Database is known as Data Mining. Association rule mining,

46 www.erpublication.org
 Churn Prediction in Telecom Industry Using R

clustering, classification and regression forms the four management that was assumed to determine the customer
techniques used by data mining. turnover is called as Churn management. (Hadden, Tiwari,
Roy and Ruta, 2007). Customer movement from one
In Data mining new rules and patterns can be discovered by provider to another in telecommunication industry is called
the system known as discovery oriented and system can also customer churn and the operators process to retain profitable
check the users hypothesis called verification oriented. It customers counted as churn management (Berson, Smith &
helps in taking knowledge-driven decisions and for predicting Thearling, 2000) [13].
the future trends of the business.
2.5 Data Set Used
2.2. J48 Decision Tree Technique
J48 construction is like a flow- chart. A test applied on an The attributes in our data are taken from Orange Database.
attribute is denoted by internal node, its effect is denoted by a
Table I: Orange Dataset Attributes
branch and class labels are presented by leaf- nodes. Process
divided in two levels, one is Division of root is recursively
S.No. Attribute name
based on selection of attribute for all training examples at the
tree construction and second is that the noise or outliers 1 State
branches are identified and removed by Tree pruning. Rules 2 Account. Length
can be classified from the tree. If-then statement is used to 3 Area. Code
represent the knowledge. For each path from root to a leaf one 4 Phone
rule is created.
5 Int .l .Plan
Here we use J48 for churn dataset. The attribute whose value 6 VMail.Plan
has to be predicted is known as dependent variable. Its value 7 VMail.Message
is decided by value of other attributes. These attributes that 8 Day.Mins
predict the value of the dependent variable are known as
9 Day.Calls
independent variables.
10 Day.Charge
2.3. Tool Used: A Revolution Analytics Tool - R 11 Eve.Mins
In the past few years, the fast emerging requirements from 12 Eve.Calls
both academia and industry has helped R programming 13 Eve.Charge
language to emerge as one of the necessary tool for 14 Night.Mins
visualization, computational statistics and data science. R is 15 Night.Calls
most popular in field of data science and important in Finance
16 Night.Charge
and analytics- driven companies.
R virtually consists all the possible statistical models, data 17 Intl.Mins
manipulation and charts that could ever be required by a 18 Intl.Calls
modern day scientist. One can easily use the best reviewed 19 Intl.Charge
methods from leading researchers in field of Data Science 20 CustServ.Calls
without any cost. It provides a large collection of graphical
21 Churn.
and statistical techniques, consisting of modelling (linear and
non-linear), statistical tests, time-series, classification,
clustering, etc.
R helps in representing complex data as beautiful and unique III. ALGORITHM AND LIBRARIES USED
data visualizations. Evaluation of result in R is very much
easier as we do not have to remember any clicks or steps, it is 3.1. J48 Algorithm
simply a programming language designed specifically for J48 (formula, data, subset, control= Weka_control ())
Predict is a generic function for predictions from the results of
data analysis that also has the capability to use mix and match
model fitting functions.
models for best results.
As R is supported by a large community worldwide, solution 3.1.1. Steps:
to the errors and code is available freely. Its source code is Step 1. A flow-chart-like tree structure. Internal node denotes a
written in C, Fortran and R. R is easily extensible through test on an attribute. Outcome of the test is represented by
functions and extensions, and the R community is noted for its Branch. Class labels are represented by Leaf nodes.
active contributions in terms of packages. R is an open source Step 2. Decision tree generation comprised of two phases.
and can be extended easily as individuals using it can Tree construction: At start, root contains all the training
contribute in its growth. Dynamic and static graphics are examples. Tree pruning: Branches that reflect noise and outliers
available through additional packages. R can easily deal with are identified and removed.
complex and large datasets. Step 3. Decision tree is used to classify an Unknown sample.
Attribute values of the sample are tested against the decision
The libraries and packages of R that are being used in this
tree.
paper are: RWeka, ggplot2, rpart, rJava, class Step 4. When all samples for a given node belong to the same
2.4. Related Literature class, or there are no remaining attributes for further
partitioning then the partitioning is stopped.
Churn customer is one who leaves the existing company and
become a customer of another competitor company. The

47 www.erpublication.org
 International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869, Volume-3, Issue-5, May 2015
3.1.2. Extracting Classification Rules from Trees Table II: Description of Data Set Attribute

1. IF-THEN rules are used.

2. From root to leaf one rule is created for each path.
3. Each attribute-value pair along a path forms a conjunction.
4. The leaf node holds the class prediction.
5. Rules are easier for humans to understand.

3.2. Using rpart package

rpart (formula, data, method)

f<-rpart(Churn.~CustServ.Calls+Eve.Charge+Intl.Charge+Night.C
harge+Day.Charge, method="class", data=churn)
Package rpart is used in plotting the graphs. The functions within
rpart that are used are as follows:

3.2.1.1. Using plotcp function

Applied on the set of possible cost- complexity pruning of a tree

from a nested set. A cross- validation is already performed by rpart
on the geometric means of the Interval values of cp where pruning is
optimal. The mean and standard deviation of errors in cross-
validated prediction against each of the geometric means is stored in
cptable in f are plotted by this function. A good choice of cp for
pruning is often the leftmost value for which the mean lies below the
horizontal line. 4.4. Summary of Data set

Table III: Summary of Dataset

3.3. Using Plot function

plot(Churn. ~., data = churn, type = "c")

lines(Churn.~ Day.Charge,type="l")
In plot function, x and y axis are mentioned along with the data
source and the type of graph that is, curve, line etc.

3.4. Using ggplot2 package

qplot is the basic plotting function in the ggplot2 package. It is

familiar with plot function. It is quick plot as it produces complex
plots in mere one line, which often require several lines of code
using other plotting systems. It helps depicting more than 2
variables in a single graph with help of colors and geometrical
shapes and a lot more.

IV. EXPERIMENTAL RESULTS AND OBSERVATIONS

4.1 Reading Data Set Churn from CSV file

churn<-read.csv("C:\\Users\\Documents\\R\\win-library\\3.1\\RWe
ka\\R\\churn.csv", header=T)

4.2. Names of all the attributes

> names(churn)
[1] "State" "Account.Length" "Area.Code" "Phone"
[5] "Int.l.Plan" "VMail.Plan" "VMail.Message" "Day.Mins"
[9] "Day.Calls" "Day.Charge" "Eve.Mins" "Eve.Calls"
[13] "Eve.Charge" "Night.Mins" "Night.Calls"
"Night.Charge"
[17] "Intl.Mins" "Intl.Calls" "Intl.Charge" "CustServ.Calls"
[21] "Churn."
4.5. Decision Tree for Churn (using J48)
4.3. Description of complete data Set
m2 <- J48(`Churn.` ~ ., data = churn)
m2

48 www.erpublication.org
 Churn Prediction in Telecom Industry Using R

4.6. Classification Tree for all the Calls (using rpart )

library(rpart)
f<-rpart(Churn.~CustServ.Calls+Eve.Calls+Intl.Calls+Night.Calls
+Day.Calls,method="class", data=churn)
plot(f, uniform=TRUE,main="Classification Tree for Churn")
text(f, use.n=TRUE, all=TRUE, cex=.7)

Fig. 2 represents the classification tree for all the Calls considered in
churn Dataset. The decision is made on basis of call number and the
churn factor having values true and false.

Figure 2: Classification Tree Based on Calls

4.7. Using rpart

library(rpart)
f<-rpart(Churn.~CustServ.Calls+Eve.Charge+Intl.Charge+Night.
Charge+Day.Charge, method="class", data=churn)
plotcp(f,lty=4,col="red")
Fig. 3 represents Applied on the set of possible cost- complexity
pruning of a tree from a nested set. A cross- validation is already
performed by rpart on the geometric means of the Interval values of
cp where pruning is optimal. The mean and standard deviation of
errors in cross- validated prediction against each of the geometric
means is stored in cptable in f are plotted by this function.

Fig. 1 depicts the churn values from table formed by predicting the
values of J48 decision tree on churn parameter.

Figure 3: Plotcp function

4.8. Using Plot Function

plot(Churn. ~., data = churn, type = "c")

lines(Churn.~ Day.Charge,type="l")

Fig. 4 represents the graph of customer service calls with respect to

Figure 1: Churn value prediction churn factor that has just two values True and false

49 www.erpublication.org
 International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869, Volume-3, Issue-5, May 2015
4.11. Line Charts

4.11.1. Line Chart for Day Calls and Customer Service Calls

# convert factor to numeric for convenience

churn$Churn. <- as.numeric(churn$Churn.)
ntrees <- max(churn$Churn.)
# get the range for the x and y axis
xrange <- range(churn$Day.Calls)
yrange <- range(churn$CustServ.Calls)
# set up the plot
plot(xrange, yrange, type="n", xlab="Day.Calls (num)",
ylab="CustServ.Calls(num)" )
colors <- rainbow(ntrees)
linetype <- c(1:ntrees)
plotchar <- seq(15,15+ntrees,1)
# add lines
for (i in 1:ntrees) {
Figure 4: Churn and Customer Service Calls tree <- subset(churn, Churn.==i)
lines(tree$Day.Calls, tree$CustServ.Calls, type="b", lwd=1.5,
Fig. 5 represents the graph of churn factor with respect to all the lty=linetype[i], col=colors[i], pch=plotchar[i]) }
states. The number of churns can easily be observed state wise in # add a title and subtitle
the graph. title("Churn", "line plot")
# add a legend
legend(xrange[1], yrange[2], 1:ntrees, cex=0.8,
col=colors,pch=plotchar, lty=linetype, title="Tree")

Fig. 7 shows the line chart of Day calls and Customer Service calls
using numbers as range and considering the Churn factor. The
number of churns increase with the increase in customer service
calls.

Figure 5: Churn with respect to different States

Fig. 6 represents the graph of churn factor with respect to

International plan.

Figure 7: Line Chart Day Calls and Customer Service Calls

4.12. Using qplot function

4.12.1. qplot(Day.Calls, CustServ.Calls, data =

churn,colour=Churn.)

Fig. 8 shows us relativity between customer service calls and Day

calls with respect to the churn factor that is represented by two
colors. By the color in the graph we see that churners are more in
high number of customer service calls. Blue color represents the
customers who churned.

Figure 6: Churn and International Plan

50 www.erpublication.org
 Churn Prediction in Telecom Industry Using R

Figure 10: Relativity In Customer Service Calls, Day Calls and

Churn factor
Figure 8: Qplot with three parameters
4.12.4. qplot(Day.Calls,CustServ.Calls,
4.12.2. qplot(Day.Calls,Night.Calls, data = churn,geom =
data=churn,facets=Churn.~Area.Code)
c("point", "smooth"),color=Churn.)

Fig. 9 shows the relativity in number of night calls and day calls. We Fig. 11 shows the relativity in number of customer service calls and
observe that they are relatively dense in the same area. Whereas, the day calls on the subset of Data. The third parameter Churn factor
third parameter Churn factor represented by the color shows that and fourth is Area Code. The facets are representing third and fourth
True churns are less in number. The line in the graph is the smoother parameter. We can observe the churns in particular area code with
that depicts the trend followed by the data in graph. Here it depicts respect to number of day calls and customer service calls.
the relatively same number of night calls and Day calls.

Figure 11: Relation of Churn in Area code w.r.t. Calls

4.12.5. dsc<- churn[sample(nrow(churn), 100),]

qplot(Day.Calls,Churn., data = dsc,geom = c("point",

"smooth"),color=State)

Fig. 12 shows the graph of day calls and churn factor on the subset
of Data. The third parameter state represented by the color shows the
churns in various states.
Figure 9: Relativity in Night Calls and Day Calls with churn
factor using smooth curve

4.12.3. dsc<- churn[sample(nrow(churn),100), ]

qplot(Day.Calls,CustServ.Calls, data = dsc, geom = c("point",

"smooth"),color=Churn.)

Fig. 10 shows the relativity in number of customer service calls and

day calls on the subset of Data. The third parameter Churn factor
represented by the color. The smooth lines in the graph show clearly
that Churns are more in case of high customer service calls. Figure 12: State wise Churn factor
Whereas, they dont vary much with the day calls. 4.12.6. qplot(Area.Code,Night.Mins, data=dsc)

51 www.erpublication.org
 International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869, Volume-3, Issue-5, May 2015
4.12.9. qplot(Night.Calls, data = dsc,geom =
fig. 13 shows the relativity in Night minutes and Area code. We see "histogram",fill=Int.l.Plan)
that there is no calls made in Area code between 425 and 500.

Fig. 16 shows the histogram, color is done using the Churn factor
and represents Night Calls.

Figure 13: Relativity in Night min and Area code

4.12.7. qplot(Day.Calls,Night.Calls, data = churn, alpha=I(1/2))

Fig. 14 shows the use of alpha filter that shows transparency. It

shows where the majority of points lie in the graph.
Figure 16: Histogram for Night Calls

V. CONCLUSION
The proposed research has used data mining technique and R
package to predict the results of churn customers on the benchmark
Churn dataset available at http://www.sgi.com/tech/mlc/db/ and
http://www.dataminingconsultant.com/data/churn.txt. It has
evaluated, the number of churns using the classification technique
J48 tree. The R tool has represented the large dataset churn in form
of graphs which depicts the outcomes vividly and in a unique pattern
visualization manner. The Churn Factor is used in many functions to
depict the various areas or scenarios when the churn rate is high. The
study predicts that there is a huge deviation in graph of churners
when customer service calls are measured. The graphs are made
taking churn factors as the deciding parameters. Graphs represent
Figure 14: Alpha Filter the different ways of observing the number of churners from the
dataset. Once the root area is recognized the steps can be taken by
4.12.8. qplot(Day.Calls, data = dsc,geom = Telecom Company to improve their services and retain their old
"histogram",fill=Churn.) customers from churning

Fig. 15 shows the histogram, in which color is done using the Churn
factor and represents Day REFERENCES
Calls.

[1] Applying Data Mining Techniques in Telecom Churn Prediction ,

International Journal of Advanced Research in Computer Science and
Software Engineering, N.Kamalraj Dr.A.Malathi , Department of
Computer Technology PG and Research Department of Computer
Science, Dr. SNS Rajalakshmi College of Arts and Science Govt. Arts
College.
[2] Anshul Goyal and Rajni Mehta, Assistant Professor,CSE. Performance
Comparison of Nave Bayes and J48 Classification Algorithms.
[3] R Data: http://cran.r-project.org/
[4] Vladislav Lazarov, Technische Universitt Mnchen and Marius Capota,
Technische Universitt Mnchen, Churn Prediction,
[5] Prof S. Chandrasekhar, Predicting The Churn In Telecom Industry.
[6] Dr. Rajni Jain, Introduction to Data Mining Techniques.
[7] Marcin Owczarczuk, Churn models for prepaid customers in the cellular
telecommunication industry using large data marts, Expert Systems
with Applications, 37 (2010) 47104712.
[8] Gary M. Weiss, Data Mining in the Telecommunications Industry,
Data Mining and Knowledge Discovery Handbook, 2005 - Springer
[9] Data Mining And Business Analytics with R, Johannes Ledolter, Dept.
of Management Studies, University of IOWA.
Figure 15: Histogram for Day Calls

52 www.erpublication.org
 Churn Prediction in Telecom Industry Using R

[10] Predicting Customer Churn in Mobile Telephony Industry Using

robabilistic Classifiers in Data Mining ,Clement Kirui1, Li Hong2,
Wilson Cheruiyot3 and Hillary Kirui4.
[11] Applying Data Mining to Insurance Customer Churn Management,
Reza Allahyari Soeini and Keyvan Vahidy Rodpysh
[12] Han and Kamber. Data Mining: Concepts and Techniques. Second
Morgan Kaufman Publisher, 2006
[13] A Proposed Churn Prediction Model, Essam Shaaban, Yehia Helmy,
Ayman Khedr, Mona Nasr,Department of Information Systems,
MUST University, 6 October, Cairo, Egypt and Department of
Information Systems, Helwan University, Cairo, Egypt.
[14] Data Mining Applications in Customer Churn Management, Sahand
KhakAbi, Industrial Engineering Department, Iran University of
Science and Technology, Tehran, Iran, Mohammad R. Gholamian,
Morteza Namvar.
[15] (IJACSA) International Journal of Advanced Computer Science and
Applications, Churn Prediction in Telecommunication Using Data
Mining Technology, Rahul J. Jadhav of Bharati Vidyapeeth Deemed
University,Pune, Yashwantrao Mohite of Institute of Management,
Karad and Usharani T. Pawar2, Shivaji University Kolhapur.
[16] International Journal of Computer Applications (0975 8887) Volume
42 No.20, March 2012, Applications of Data Mining Techniques in
Telecom Churn Prediction, V.Umayaparvathi Department of
Computer Science DDE, Madurai Kamaraj University, K. Iyakutti
School of Physics Madurai Kamaraj University Madurai, Tamilnadu,
India.
[17] Yu-Teng Chang, Applying Data Mining To Telecom Churn
Management, IJRIC , 2009 67 77.
[18] Telecommunication Subscribers Churn Prediction Model Using
Machine Learning, Saad Ahmed Qureshi, Ammar Saleem Rehman,
Ali Mustafa Qamar, Aatif Kamal Department of Computing, (NUST),
Islamabad, Ahsan Rehman, Business Analytic Consultant, IBM .
[19] Goran Kraljevic, Sven Gotovac, Modeling Data Mining Applications
for Prediction of Prepaid, Churn in Telecommunication Services.
[20] Bong-Horng Chu, Ming-Shian Tsai and Cheng-Seen Ho, Toward a
hybrid data mining model for customer retention, Knowledge-Based
Systems, vol. 20, 2007, pp. 703718.
[21] X. Hu, A Data Mining Approach for Retailing Bank Customer
Attrition Analysis, Applied Intelligence, vol. 22, 2005, pp. 4760,
Springer.
[22] Data Mining in the Telecommunications Industry, Gary M. Weiss,
Fordham University, USA.
[23] Customer Churn Prediction for Telecom Services, Utku Yabas,
Turker Ince, School of Engineering and Computer Science, Izmir
University of Economics, Izmir, Turkey, Hakki Candan Cankaya,
Computer Science and Engineering Department, Southern Methodist
University Dallas, Texas, 2012 IEEE 36th International Conference on
Computer Software and Applications.
[24] The Study on Feature Selection in Customer Churn Prediction
Modeling, Yin Wu, Jiyain Qi, School of Econ+omic and
Management, Beijing University of Posts and Telecommunications,
Beijing, China, Chen Wang, IBM China Research Laboratory, IBM,
Beijing, P.R. China.
[25] Predicting Customer Behavior in Telecommunications, L. Yan, R.H.
Wolniewicz, R. Dodier, IEEE Intelligent Systems, IEEE Computer
Society.
[26] Wikipedia www.wikipedia.com.
[27] Elegant Graphics for Data Analysis, Wickham, H.
http://www.springer.com/978-0-387-981040-6.
[28] Telecommunication Subscribers Churn Prediction Model Using
Machine Learning , Saad Ahmed Qureshi, Ammar Saleem Rehman,
Ali Mustafa Qamar, Aatif Kamal

Ms.Manpreet Kaur, Student, Guru Gobind Singh Indraprastha

University, Institute Of Information Technology & Management, Mca,
Trainee At Niit Technologies, Delhi , India

Dr. Prerna Mahajan, ,Professor, Department Of Computer

Science,Guru Gobind Singh Indraprastha University/ Institute Of
Information Technology & Management, Delhi/India

53 www.erpublication.org