See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/324591210

Predictive Analytics for Increased Loyalty and Customer Retention in

Telecommunication Industry

Article  in  International Journal of Computer Applications · April 2018

DOI: 10.5120/ijca2018916734

CITATIONS READS

2 1,941

3 authors:

Oladapo Kayode Omolola Omotosho

Baptist Mission Schools of the Nigerian Baptist Convention Covenant University Ota Ogun State, Nigeria
11 PUBLICATIONS   5 CITATIONS    29 PUBLICATIONS   113 CITATIONS   

SEE PROFILE SEE PROFILE

Oladapo Adeduro

1 PUBLICATION   2 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Effect of hydro-ethanolic extract of laportea aestuans on benign prostatic hyperplasia in male wistar rats View project

Pluvial Flood Risk Assessment View project

All content following this page was uploaded by Oladapo Kayode on 02 July 2018.

The user has requested enhancement of the downloaded file.

 International Journal of Computer Applications (0975 – 8887)
Volume 179 – No.32, April 2018

Predictive Analytics for Increased Loyalty and Customer

Retention in Telecommunication Industry
Oladapo K. A. Omotosho O. J. Adeduro O. A.
Department of Computer Department of Computer Department of Computer
Science, School of Computing Science, School of Computing Science, School of Computing
and Engineering (CES) and Engineering (CES) and Engineering (CES)
Babcock University, Ilisan- Babcock University, Ilisan- Babcock University, Ilisan-
Remo, Ogun State, Nigeria Remo, Ogun State, Nigeria Remo, Ogun State, Nigeria

ABSTRACT 2. LITERATURE REVIEW

Literature has indicated that to engage a new customer cost at
least 6 – 10 times higher than retaining the existing ones. The 2.1 Customer Loyalty
competitive nature of the telecommunication industry has Loyalty is seen as a concept which has its base from the
made customer retention to be a crucial responsibility for customer behaviour theory and reflects what customer may
telephone services provider. Since customer retention is a portray to services, activities or brand (Boohene and
vital element for every establishment to be conscious of in Agyapong, 2011). The concept has been identified as part of
retaining loyal customers, so also is the ability to perfectly main sources of industry scores and competitive advantage.
predict customer retention is very necessary. Customer Industry growth and profit are primarily stimulated by
retention prediction models are highly needed by the customer's loyalty (Heskett, Jones, Loveman, Sasser, &
telecommunication industry to efficiently manage the Schlesinger, 2008). Customer loyalty is the willingness of a
retention of existing customers. This paper proposes a logistic customer to retain his/her relationship with a certain industry,
regression model to predict customer retention in the product or services (Khan, 2012). There are two dimensions
telecommunication industry. The results indicate that logistic to loyalty; behavioural and cognitive. Behavioural include the
regression can predict customer retention with the accuracy of intention of a customer to repurchase, to change and give
95.5%. Furthermore, it was observed that when billing issues mainly to the industry, while cognitive is the preference over
are resolved it is more likely to retain customer while value- an industry, advocating for the industry and ability to pay
added service and short message service issues are associated more for services (Taylor, 2007). Enhanced customer loyalty
with the likelihood of exhibiting customer retention. has a lot to do with industry profitability.

General Terms 2.2 Customer Retention

Predictive Modelling, Predictive Analytics, Data Mining Customer retention is the continuous maintenance of a trading
Techniques, Information System. relationship with customers over a longer period of time. It is
the total number of customers dealing with an industry at the
Keywords end of the financial year in respect to those who were active at
Prediction, Predictive Analytics, Loyalty, Customer Retention the beginning of the year (Dawkins and Reichheld, 1990).
Customer retention is very important for the existence of an
1. INTRODUCTION industry because it drives profitability. Industry is expected to
The telecommunication industry has more competition than increase retention activities mostly to all loyal customers.
every other industry which gives the customers more choices. Traditional product-oriented industry aims at customer
To acquire a new customer costs at least 6 – 10 times more retention as major objectives by putting in more effort to be a
than retaining the existing ones (Pogol, 2007). To every more customer-focused element in line with the principles of
industry the cost acquiring new customers takes the large sum marketing (Kapai and Moronge, 2015). When an industry
of the administrative expenses; consequently, customer retains only 5% of customers, such industry's profit gear up by
loyalty and retention turns into the major factors to be 25% to 125% (Alshuredi & Alkurdi, 2012). Larivière and Van
considered in maintaining growth knowing fully well that den Poel (2005) believes turning customer relationship to be
industries in a competitive market like telecommunication more personal has a lot to do with customer retention.
majorly depends on the ever-growing profits that come from
the existing loyal customers. Therefore efforts are to be made 2.3 Related Work
to increase loyalty and retain customers. It is of high benefit to Different efforts have been put in place in building an
the industry to build a more effective customer retention effective prediction model for customer retention using
prediction model. different techniques. For a better understanding of how
several studies have built their prediction models, this article
This paper suggests logistic regression model design, a good reviews few as shown in table 1.
model from acquired customer data to predict retention. The
remaining part of the paper is organized as follows. Section 2
reviews literature in relation to customer loyalty and retention.
Section 3 proposes a model for customer retention and section
4 presents the results of the acquired data set, finally, section
5 provides the conclusion.

43
 International Journal of Computer Applications (0975 – 8887)
Volume 179 – No.32, April 2018

Table 1: Related literature about customer retention in Kapai and Moronge (2015) applied linear regression in a
telecommunication industry mobile communication industry in order to explore how
customer satisfaction, customer care service, sales promotion
Author Data set Predictive Method and tariff structure influences customer retention.
Kapai and Airtel Kenya Linear Regression Adebiyi, Oyatoye and Mojekwu (2015) built a prediction
Moronge model for Nigeria’s telecommunicating industry by using
(2015) Markov Chains and a time-homogenous model to predict
Adebiyi, Subscriber Markov Chain (Time- customer churn and retention rates.
Oyatoye and Database homogenous model) From the review of the literature, it can be concluded that any
Mojekwu of the predictive models can predict customer retention in the
(2015) different industry. This paper proposes a logistic regression
Sharma and UCI Repository Neural Network approach to predict customer retention in the
Panigraha of Machine telecommunication industry.
(2011) Learning
Database
3. THE PROPOSED SSLR MODEL
The main objective of industry offering services should be
Su, Coopre, Customer Clustering (K-means) customer satisfaction which in one way or the other bring
Probinhei and Analytic record them back, but if the customer is loyal, he/she will surely
Jordan (2015) come back (Kapai and Moronge, 2015). SSLR Model is
proposed based on the outcome of the variable performance
Flaherty and Publishing Design science - called customer retention. Customer satisfaction is a vital
Heavin (2015) Company Predictive Decision element for those in the services industry (Lewis, 2004). The
Support System SSLR model has four qualities namely "S" for Service
(PDSS) Quality, "S" for Customer Satisfaction, "L" for Customer
1) survival analysis Loyalty and "R" for Customer Retention.
(Cox’s regression), Service Quality is used in measuring billing issues, value
(2) logistic regression added services issues and Short Message Service (SMS)
and issues, out of various numbers of consumer complaints.
Customer Satisfaction measures the industry performance
(3) decision Trees which occurs when the quality of service exceed customer’s
(Chi-squared expectation, with this formula:
Automatic Interaction
Detector) CS = SQ > CE
Where CS is customer satisfaction, SQ is quality of service
and CE is customer expectation. This shows that customer
satisfaction can be influenced by two different management

Billing Value Added Service Short Message Service

Service Quality

SQ > CE

Customer Satisfaction

CL = f (CS)

Customer Loyalty

CR = f (SQ + CS + CL)

Customer Retention

Figure 1: Proposed Model for Customer Retention

44
 International Journal of Computer Applications (0975 – 8887)
Volume 179 – No.32, April 2018

(i) customer expectation management and (ii) industry in a data. usage is

performance management. substandard to
performance base
Loyalty is used to measure customer satisfaction that is
on the acquired
customer loyalty is a function of customer satisfaction,
data.
mathematically:
Support Vector It can be used for It is very hard to
CL = f (CS)
Machine non-linear decision understand and
Where CL is customer loyalty and CS is customer making. explain without
satisfaction. resorting to the
needed
Retention indicates customer retention as the outcome of the mathematics.
industry performance against three variable namely Service
Quality (from the perspective of billing, value-added service Cluster Models Previous knowledge Struggles with
and SMS), Customer Satisfaction and Customer Loyalty. about behaviour is large quantities of
not necessary. data both in terms
CR = f ( SQ + CS + CL) of the outcome
Where CR is customer retention, SQ is service quality, CS is data and predictor
customer satisfaction and CL is customer loyalty. variables.

3.1 Comparison of Predictive Model Source: Finlay (2014)

Predictive Analytics is the most common data mining 3.2 Dataset
technique that uses mode to predict what might happens in a
This paper has used the consumer complaint from the
particular incident. Alan (2009) described predictive
Nigerian Communication Commission statistics and reports.
modelling as an aspect of statistics and data mining technique
The consumer complaints dataset deals with telephony service
used in getting information from data and using such
provider's customer and the data apt to the quality of service
information to predict events. It comprises of a number of
they receive. Consumers have the choice of telephony service
predictors, variables that possibly will assist in future events.
provider or companies providing them with a telephony
In the use of predictive modelling, the collection of data is the service. When these consumers receive good quality of
first approach, followed by the formulation of a statistical service, they are said to be satisfied thus increase loyalty and
model, after which prediction is made and the model is then retained which in the long run increase profit. The
revised as more data show-up. According to Nishchol and telecommunication industry concerned here has used three
Sanjay (2012), the steps involved in predictive models are the complaints with the highest frequency out of the various listed
definition of a project, analysis of source data (exploration), unresolved complaints by telecommunication consumer and
preparation of data, the building of a model, application of generated a list of precise records related to the study. The
model result (deployment) and management of model. logistic regression is implemented on SPSS, the dataset
contains three variables worth of information about 18,711
The Predictive model exists in different shapes and sizes and customers, along with the indication of whether or not the
there are more than enough that can be used to create a model. consumer is retained or ported. The variables are billing issue,
A review of the strength and weakness of the most common value-added services issue and short message service issue.
types of predictive model is presented in table 2: The data was streamlined to 202 customers based on the
Table 2: Comparison of Predictive Model incoming and outgoing porting data for the third quarter 2017.

Types of Strength Weakness 3.3 Logistic Regression

Predictive Regression is the measure of the functional relationship
Model between two or more variables in terms of the original units of
the data. There are different types of regression namely,
Linear Model It is easy to It creates accurate simple linear regression, multiple linear regression. Among
understand and predictions if and many types of regression, the most common in data mining is
explain. It can also only if the logistic regression which is popularly called a logit model is
be adjusted to relationship in the used to model categorical outcome variable (dependent)
prevent overfitting. data is linear, where the log odds of the outcome (dependent variable) is
Used in though it can be modelled as a linear combination of the predictor variable
benchmarking, overcome. (independent variable). Each predictor is given a coefficient
serves as a standard ‘b' which measures its independent contribution to variations
for other models in the dependent variable which can only take one of the two
Decision Trees Ability to segment a Most algorithms values: 0 or 1. Logistic regression was used because having a
(Classification population into for developing dichotomous outcome variable violates the assumption of
and Regression) smaller and smaller decision trees are linearity in the normal regression. There are three types of
segments. not very effective logistic regression namely:
in using data. - Binary logistic regression model: used to model a binary
Easy to understand
and explain to a non- response
technical audience. - Ordinal (ordered) logistic regression model: used to model
Neural Ability to take into It is complex. an ordered response
Networks consideration non-
linear characteristics Performance in

45
 International Journal of Computer Applications (0975 – 8887)
Volume 179 – No.32, April 2018

- Nominal (unordered) logistic regression model: used to Table 5 shows the contribution of each independent
model a multilevel response with no order. variable to the model and its statistical significance.
This paper adopts the use of a binomial logistic regression Variables in the Equation
which predicts the chances that an observation will fall into
one of two categories of a categorical dependent variable with B S.E. W d Si Exp(B) 95%
respect to one or more continuous independent variables. In al f g. C.I. for
this case, the binomial logistic regression was used to d EXP(B)
understand whether customer retention can be predicted based Lo Up
on billing, value-added service and short message service we per
issues. Where customer retention is the dependent variable r
measured on categorical scale "yes" or "no" and three
independent variables: "billing issue", "value added service St Billing_I 23. 4465. .0 1 .9 12026312 .00 .
issue" and "short message service issue" e ssue(1) 21 886 00 96 891.738 0
p 0
4. RESULT AND DISCUSSION 1a
The analysis part of the generated model displays information VAS_Iss 19. 1271 .0 1 .9 21700406 .00 .
about the logical regression. ue(1) 19 0.132 00 99 8.237 0
5
Table 3 shows the model summary in order to depicts how
much variation is in the dependent variable. SMS_Iss .00 1441 .0 1 1. 1.000 .00 .
ue(1) 0 1.936 00 00 0
0
Model Summary Constant - 6793. .0 1 .9 .000
Step -2 Log Cox & Snell Nagelkerke R 21. 852 00 98
likelihood R Square Square 20
3
1 55.293a .667 .893
a. Variable(s) entered on step 1: Billing_Issue, VAS_Issue,
a. Estimation terminated at iteration number 20 because SMS_Issue.
maximum iterations has been reached. Final solution cannot
be found.
In this case, it was noted that billing, value-added service and
short message service issues contributed significantly to the
The Nagelkerke R Square is 0.893 which indicates there is a prediction (p = 0.996, 0.999 and 1.000 respectively.)
strong relationship of 89.3% between the predictors (billing,
value-added service and short message service issues) and the 5. CONCLUSION AND FUTURE WORK
prediction (customer retention). A logistic regression was performed to ascertain the effects of
billing, value-added service and short message service issues
Table 4 represents the category prediction that estimates
on the likelihood that customers will be retained. The logistic
the probability of an event.
regression model was statistically significant. The model
Classification Tablea explained 89.3% (Nagelkerke R Square) of the variance in
customer retention and correctly classified 95.5% of issues.
Observed Predicted When billing issues are resolved it is more likely to retain
Customer Percentage customer while value-added service and short message service
Retention Correct issue are associated with the likelihood of exhibiting customer
retention. Customer retention prediction is very important in
No Yes telephone services providers in developing countries. So as to
be competitive in the telecommunication industry, telephone
Step Customer No 112 0 100.0 service providers are expected to be able to predict customer
1 Retention retention through their complaint, therefore assist in taking
Yes 9 81 90.0
proactive steps in retaining the valuable loyal consumers.
Overall 95.5 Therefore to building an efficient model for customer
Percentage retention is an issue for those both in the industry and
academic world. This article suggests that data mining
a. The cut value is .500
technique can be a bright answer for customer retention and
increased loyalty. The final model summary is this article
In the study, 90% were correctly classified for customer indicated that the model gives more than 95.5% accuracy for
retention and 100% for the no customer retention and overall the prediction of customer retention.
95.5% were correctly classified. This is a considerable
improvement on the 100% correct classification with the For future work, other consumer complaint issues can be
constant model so it indicates that the model with predictors is considered, along with other prediction techniques such neural
a significantly better model. networks and two prediction techniques can be combined to
develop a hybrid model. Finally, the present methodology of
customer retention can be tested for other industry like airline,
insurance and banking.

46
 International Journal of Computer Applications (0975 – 8887)
Volume 179 – No.32, April 2018

6. ACKNOWLEDGMENTS [9] https://www.ncc.gov.ng/stakeholder/statistics-

The authors would like to thank Mr. Akinsola J.E.T for his reports/consumer-complaints
effort in making this paper work a reality. Also appreciate the [10] Kapai R. and Moronge, M. (2015). Determinants of
anonymous reviewers for their contributions in improving this effective customer retention in the mobile
paper work. communication industry in Kenya. The strategic journal
of business and change management. Vol. 2, Iss. 2 (108),
7. REFERENCES pp 1621 – 1672, Nov 2, 2015.
[1] Adebiyi S.O, Oyatoye E.O & Mojekwu E.N (2015).
Predicting Customer Churn and Retention Rates in [11] Khan, (2012). Impact of Customer Satisfaction and
Nigeria’s Mobile Telecommunication Industry Using Customers Retention on Customer Loyalty. International
Markov Chain Modelling, ActaUniv. Sapientiae, Journal of Scientific and Technology Research, 1(2),
Economics and Business, 3 (2015) 67–80, DOI: 106-110.
10.1515/auseb-2015-0004
[12] Larivière, B., & Van den Poel, D. (2005). Predicting
[2] Alan M. (2009). “Introduction to Forecasting Methods customer retention and profitability by using random
for Actuaries.” Forecasting & Futurism Newsletter, forests and regression forests techniques. Expert Systems
September, pp 6-9. with Applications, 29(2), 472-484.
https://soa.org/Library/Newsletters/Forecasting-
[13] Lewis, M. (2004). The influence of loyalty programs and
Futurism/2009/September/ffn-2009-iss1.pdf
short-term promotions on customer retention. Journal of
[3] Alshuredi & Alkurdi, (2012) in Kapai R. and Moronge, Marketing Research, 41(3), 281-292.
M. (2015). Determinants of effective customer retention
[14] Nishchol M and Sanjay S. (2012). Predictive Analytics:
in the mobile communication industry in Kenya. The
A Survey, Trends, Applications, Opportunities &
strategic journal of business and change management.
Challenges. International Journal of Computer Science
Vol. 2, Iss. 2 (108), pp 1621 – 1672, Nov 2, 2015.
and Information Technologies, Vol. 3 (3) , 2012, 4434-
[4] Boohene, R. & Agyapong, G.K.Q. (2011). Analysis of 4438
the antecedents of customer loyalty of
[15] Pogol, G. (2007). Tips for Cost-Effective Customer
telecommunication industry in Ghana: The case of
Retention Management, retrieved from
Vodafone (Ghana). International Business Research,
www.crm2day.com/library/docs/50577-0.pdf.
4(1), 229-240.
[16] Sharma A. & Panigrahi K. (2011). A Neural Network
[5] Brian O’Flaherty & Ciara Heavin (2015). Positioning
based Approach for Predicting Customer Churn in
predictive analytics for customer retention, Journal of
Cellular Network Services. International Journal of
Decision Systems, 24:1, 3-18, DOI:
Computer Applications (0975 – 8887) Volume 27– No.
10.1080/12460125.2015.994353
11, August 2011
[6] Dawkins, P.M. and Reichheld, F.F. (1990). Customer
[17] Su J., Cooper K., Robinson T., and Jordan B. (2015).
retention as a competitive weapon, Directors & Board,
Customer Retention Predictive Modeling in HealthCare
Summer, pp. 42–47.
Insurance Industry. BlueCross BlueShield of Florida,
[7] Finlay S. (2014). Predictive Analytics, Data Mining and Jacksonville, FL.
Big Data. Palgrave Macmillan, UK
[18] Taylor S., Hunter G., and Longfellow T. (2006). Testing
DOI.10.1057/9781137379283
on expanded attitude model of goal directed behavior in
[8] Heskett, J.L; Jones, T.E; Loveman, G.W; Sasser, W.E & a loyal context. Journal of Consumer Satisfaction,
Schlesinger, L.A (2008). Putting the Service Quality Dissatisfaction and Complaining Behaviour. Vol. 19 p
Chain to Work, Harvard Business Review, July-August 18-33

IJCATM : www.ijcaonline.org
47

View publication stats