You are on page 1of 74

UNIVERSITY SCHOOL OF MANAGEMENT

STUDIES
ACADEMIC YEAR: 2021-22

PROJECT REPORT:
PREDICTIVE ANALYSIS ON CUSTOMER
CHURNING IN TELECOM INDUSTRY
(WIRETEL)
Post Graduate Diploma In Data Analytics

(SEMESTER – I)
Submitted To: Submitted by:

Mr. Santanoo Pattnaik Tanaya Srivastava

(University School Of Management (02016640621, PGDDA 1st Sem)


Studies, Guru Gobind Singh Indraprastha Akansha Bansal

University, Dwarka, New Delhi) (02716640621, PGDDA 1st Sem)


CERTIFICATE

This is to certify that the project report entitled “PREDICTIVE ANALYSIS ON


CUSTOMER CHURNING IN TELECOM INDUSTRY (WIRETEL)”, is the bona fide
research carried out by Ms. Akansha Bansal and Ms. Tanaya Srivastava, students
of Post Graduate Diploma in Data Analytics, Guru Gobind Singh Indraprastha
University, Dwarka, New Delhi, during the session 2021-22, in partial fulfilment of
the requirements for the award of the diploma of Post Graduate Diploma in Data
Analytics and that the dissertation has not formed the basis for the award previously
of any degree or diploma to the best of my knowledge and belief.

Dr. Santanoo Pattnaik

University School Of Management Studies,

Guru Gobind Singh Indraprastha University


DECLARATION

We hereby declare that the project report entitled – “PREDICTIVE ANALYSIS ON


CUSTOMER CHURNING IN TELECOM INDUSTRY (WIRETEL)”, is a record of original
work done by us under the guidance of Prof. Santanoo Pattnaik and is being
submitted as main project of Ist Semester in Post Graduate Diploma in Data Analytics
to University School of Management Studies, GGS Indraprastha University, Dwarka
as an authentic record of our genuine work done. The results embodied in this
research has not been submitted to any other university or institute for the award of
any degree or diploma.

Date: Tanaya Srivastava

(02016640621, PGDDA 1st Sem)

Place: Akansha Bansal

(02716640621, PGDDA 1st Sem)


ACKNOWLEDGEMENT

We express our sincere indebtedness towards our guide Prof. Shantanu Pattnaik
USMS, GGSIPU, Dwarka for his invaluable guidance, suggestions and supervision
throughout the work. Without his kind patronage and guidance the project would not
have taken shape. We would also like to express our gratitude and sincere regards
for his kind approval of the project, time to time counselling and advices.

We owe sincere thanks to all the faculty members in the department of Data analytics
and Management for their kind guidance and encouragement from time to time.

Date: 15/03/2022
TABLE OF CONTENTS

SNO. CONTENTS
1. ABOUT THE TOPIC

2. INTRODUCTION

3. CHAPTER I: EXPLORING DATASET AND DATA DESCRIPTION

4. CHAPTER II: DATA VISUALIZATION

5. CHAPTER III: DATA PROCESSING AND LINEAR DISCRIMINANT


ANALYSIS

6. CHAPTER IV: FACTORS THAT EXPLAIN CUSTOMER CHURN


• DECISION TREE
• ARTIFICIAL NEURAL NETWORK
• K-NEAREST NEIGHBOUR
• CLUSTERING
7. CHAPTER V: WHY CUSTOMERS ARE CHURNING FROM
WIRETEL?

8. CHAPTER VI: RECOMMENDATION BE TO THE TOP


MANAGEMENT TEAM OF WIRETEL
9. CHAPTER VII: DRAFTING A MORE FOCUSED RETENTION
STRATEGY FOR WIRETEL

10. SUSTAINABLE DEVELOPMENT GOALS


ABOUT THE TOPIC:
Customer acquisition and retention is a key concern for several industries and is
particularly acute in fiercely competitive and fast growth businesses. Retaining a
loyal customer is far more important than acquiring a new one, thus making
customer churn one of the critical concerns for big Telecommunication
corporations. Finding factors triggering customer churn is vital to implement
necessary remediation to pre-empt and cut back this churn. Once a company
starts losing its customers, the rapid decline in revenues makes it financially
infeasible to initiate new retention programs. Hence, predictive analysis of
customer retention is an absolute necessity in all businesses, especially in the
case of Telecommunications Industry.

The project analysis focuses on implementing machine learning (ML) algorithms


to identify potential churn customers, categorise them based upon usage patterns,
and visualize the analysis results. The predictive model demonstrates that ML
algorithms can successfully predict potential customer churn and help in devising
customer retention programmes.

1
INTRODUCTION

Telecom industry is constantly evolving and innovating. Due to the ever-increasing


competition among the corporations, an increased importance is being given to
targeted promoting methods for customer churn management. Modern day
customers expect the best services at affordable rates. In case they are
not satisfied, they quickly switch to another telecom network. Companies must
find innovative ways to predict potential customer churn in order to prosper in
such a competitive market.

Customer churn is defined as the proportion of customers who stopped using a


particular company’s products or services during a definite time frame. Telecom
industry is known to suffer from customer churn. Churn, also called attrition, is a
term used to indicate a customer leaving the service of one company in favor of
another company. Churn rate has strong impact on the life time value of the
customer because it affects the length of service and the future revenue of the
company. Telecom companies spend hundreds of dollars to acquire a new
customer and when that customer leaves, the company not only loses the future
revenue from that customer but also the resources spend to acquire that
customer. Churn erodes profitability.

Advantages and Disadvantages of Churn Rate

➢ Advantages
The advantage of calculating a company's churn rate is that it provides clarity on
how well the business is retaining customers, which is a reflection on the quality of
the service the business is providing, as well as its usefulness. If a company sees
2
that its churn rate is increasing from period to period then it understands that a
fundamental component of how it is running its business is flawed. The company
may be providing a faulty product, it may have poor customer service, or its product
may not be attractive to individuals who decided the cost is not worth the utility.

The churn rate will indicate to a company that it needs to understand why its clients
are leaving and where to fix its business. The cost of acquiring new customers is
much higher than it is to retain current customers, so as you ensure that the
customers you worked hard to attract remain as paying customers, it makes sense
to understand the quality of your business.

➢ Disadvantages
One of the limitations of the churn rate is that it does not take into consideration the
types of customers that are leaving. Customer decay is primarily seen in the most
recently acquired customers.

Perhaps your company had a recent promotion that attracted new customers. Once
this promotion was over or even if the benefit of the promotion never ended,
customers that were trying out the product may determine it's not for them,
cancelling their subscription.

The impact of losing new customers versus long-term customers is critical. New
customers are transient whereas old customers are entrenched and have enjoyed
your product and there must be a more significant reason as to why they are leaving.
A high churn rate in one period may be indicative of a high growth rate from the
previous period rather than a judgment on the quality of the business.

3
The churn rate also does not provide a true industry comparison of the types of
companies within an industry. Most new companies will have a high acquisition rate
as new people try the business, but they will also have a higher churn rate as these
new clients leave.

A company that is mature and has been around for decades will have a low churn
rate as its clients are established but its acquisition rate will also be lower.
Comparing the churn rates of both these companies will be like comparing apples
and oranges.

Telecommunications Industry Churn Rates


The churn rate is a particularly useful measurement in the telecommunications
industry. This includes cable or satellite television providers, Internet providers, and
telephone service providers (landline and wireless service providers).

As most customers have multiple options from which to choose, the churn rate
helps a company determine how it is measuring up to its competitors. If one out of
every 20 subscribers to a high-speed Internet service terminated their subscriptions
within a year, the annual churn rate for that Internet provider would be 5%.

How do you calculate churn rate?


To calculate the churn rate, choose a specific time period and divide the total number
of subscribers lost by the total number of subscribers acquired, and then multiply
for the percentage.

For example, say in a quarter you acquired 100 new subscribers but you lost 12
subscribers, your churn rate would be (12 / 100) x 100 = 12%.

4
You can also calculate the churn rate by dividing the number of subscribers lost in
a period by the total number of subscribers at the beginning of that period.

What is a good churn rate?


Ideally, a churn rate of zero would be the best churn rate, as that would indicate a
business is not losing any subscribers; however, that is never the reality. A business
will always lose subscribers for one reason or another.

In this case, it is important to compare the churn rate of the business to its industry's
average churn rate, taking into consideration if the business is new or mature.
Knowing an industry's churn rate versus that of the business is the only way to
understand if a churn rate is acceptable or poor. Every industry has a different
business model and, therefore, will have different acceptable churn rates.

What does a high churn rate mean?


A high churn rate indicates that a business is losing significant customers, certainly
more than it is bringing in. This would mean that the business is doing something
wrong, whether that be delivering a poor product, having poor customer service, or
a host of other negative reasons that would explain why it is losing customers fast.
A high churn rate would most likely mean a company is suffering significant losses.

Usage of Predictive Analytics to derive actionable insights

The real-time data that is being gathered from multiple sources like the call logs, call
records, network performance, live network data can be collected. We can correlate
this data with the customer preferences, usage history, complaints or calls received
to the call center, and customer segments from the Back end systems. Using
Predictive Analytics, we can predict the possible impact of the network events in
5
these events. This actionable insight will help the telco’s to avoid the potential risk
associated with the customer experience. Application of predictive statistical models
such as Logistic Regression, Market Basket Analysis, and Exploratory Data Analysis
can help to identify the best possible option.

Analyzing the data from the customer can provide a 360-degree comprehensive view
of the available pieces of information. This can help in personalized service offering
to the customer to retain them for a longer time. The data about a single customer
is interesting but may not be very actionable. By analyzing the data of all customers
telecom companies can identify trends, patterns, and conduct correlation analysis to
understand what factors drive service usage behavior and influence customer
satisfaction.

Main causes of customer churn in the telecom industry

The ideas for reducing customer churn was already mentioned in one of our blog
posts. This time, however, we are going to discuss this issue concerning the
telecommunications sector. Using data monetization to reduce customer churn in the
telecom industry offers various benefits, but let’s start with the main reasons why
exactly does it happens.

Six key factors lead customers to leave one telecommunications provider. To avoid
customer churn in the telecom industry, companies should address the below
factors.

6
➢ Service quality
If a network quality issue causes the churn, telco managers could identify the
exact cause of the issue effectively. For example, the main reasons for bad
service could be a damaged cell phone, poor reception, or communication
tower issues. To accurately track this driver, telcos should keep an eye on and
manage their dropped-call rate as well as their call completion rate.
➢ Features and content availability
It occurs in marketing or sales when a customer signs up for a plan that isn’t
best suited to their needs. Unfortunately, such a decision most often ends up
in terminating the contract. Therefore, it’s crucial to measure your customer
churn rate for specific services or bundles to identify an acquisition problem.
➢ Lower cost substitutes from competitors
Telcos are always in a state of instability where even the slightest change, such
as pricing changes or new product launches, can cause customers to try
another service provider. In addition, from the awarding of new telco licenses
introducing new competitors to higher taxes borne by the consumer, these
factors also contribute to customer churn in the telecom sector.
➢ Negative customer service experiences
Customer service has always affected customer satisfaction, regardless of
industry. This is even more visible in the telecom industry, where
communication services are a necessity. From issues reaching the call centres
to the actual resolution of the customer’s problems, keeping your customer
service issues to a minimum is the key.

7
Why you should analyze your churn frequently and accurately

Churn is a hugely influential statistic across a SaaS business. It’s the metric by which
businesses, young and old, live or die. Letting your churn rate creep higher can lead
to a number of related problems.

➢ Churn leads to higher CAC & reduced revenue

Customer acquisition cost (CAC) is the total cost of sales and marketing
required to acquire a customer.

Regardless of the strength of your marketing strategy and the effectiveness of


your approach to nurturing leads, every customer costs money to obtain. In
fact, acquiring new customers is considerably more expensive than
maintaining and upgrading existing customer relationships. The more
customers you churn, the more money you must spend to recoup the loss of
business by finding new ones.

CAC eats away at essential revenue and fundamentally works against financial
growth — the lower you can keep it, the better, and you keep CAC low by
actively fighting against churn.

➢ Analyzing churn helps you reduce it

Although you can attempt to outrun your churn by focusing on bringing in new
customers, you can only actively reduce your churn by focusing on the areas

8
where you’re losing customers. A company that is churning customers owing
to a poor pricing strategy will need a different improvement strategy than one
that is churning customers due to a dysfunctional customer support program.

Analyzing churn data

➢ Key performance indicators (KPIs) and goals


As with any form of analysis, successfully analyzing your churn requires you
to keep track of the right data. Setting the right KPI-oriented goals can help
you get a closer look at what is causing your churn.
A few KPIs helpful for churn analysis can include:

➢ Customer engagement and usage


The surest signal of impending customer churn is a consistently declining rate
of usage, which will suggest different things depending on how long your
customer had used your service before their usage rate declined.

➢ Support tickets
If your support team is receiving markedly fewer tickets than normal, that may
mean that you’re doing everything perfectly. It’s more likely, however, that it’s
a sign of customers’ lack of engagement, that they no longer feel it’s worth
the time to seek support to fix their problems with your service.

➢ Competitor pricing points


Monitor competitors’ pricing, as well as the deals they’re offering, as it’s
important to anticipate the likelihood of customers churning to those

9
competitors. Your pricing strategy should be a subject of continual re-
evaluation to ensure that you stay competitive.

➢ Likelihood to upgrade
Because high churn can be difficult to reduce, and even small increases in
monthly churn are highly damaging, it’s a good idea to maintain a KPI that will
allow you to hit back against churn preemptively. Securing customer
relationships with upsells is one way to do this. Are certain types of customers
more likely to upgrade their subscriptions than others? Is there a particular
point in your product lifecycle where upselling is effective? Target these areas
and take the time to nurture and improve those relationships. That way, even
if you’re battling churn elsewhere, you are still securing the value of your other
existing subscriptions.

➢ Customer behavioral patterns


It's unlikely that your company will just be serving one type of customer.
Different customers have different needs, and these different needs translate
to different behavioral patterns. This is where cohort analysis comes in.
Separating your customers into cohorts based on when they sign up and their
product usage behavior can be hugely helpful in analyzing, anticipating, and
preventing churn. Note similarities in customer behavior, like:

o Which customers are more likely to disengage after a given period of


time
o Which features of your product are the most highly or underutilized

10
o Which customers are at a higher risk of delinquent churn
o Which portion of customers are using your support services

From there, address strategies to re-engage with customers who are no longer
getting value out of your product. Use email marketing or direct outreach to
customers who have begun to drift away. Deliver alerts to customers whose
subscriptions are about to end or whose payment methods are out of date.

➢ Customer segmentation
To get an even firmer handle on the behavior of your customers, split them
into separate groups based on their industry, how long they've been using
your product or service, and patterns in their usage to see who’s more at risk
of churn and how to engage with them.

o Observe how customers in different industries use your product and


which customers are more churn-susceptible.
o Observe how the churn risk changes depending on how long your
customers have been using your product.
o Observe how users in differing pricing tiers operate with respect to
churn — are customers in lower tiers getting enough features to stop
them from losing engagement? Are customers in higher tiers being
charged too much and churning to better-value competitors?

11
➢ Touchpoint behavior
Customers who become frustrated with your service, with no support to turn
to, are likely to churn. Establish points of contact for support so that your
customers can get all the value they’ve paid for from your product, and you
can gain valuable information on how your product can be improved.

Why you need to analyze your churn?

If your company is looking for a solution that will allow you to get a better grip on
your churn analysis, you should select one that takes a holistic approach to the
various types of potential churn.

➢ Predict your churn


A useful solution will be able to predict the most likely causes of churn and
flag any customers at risk. For example, how much of your churn is coming
from simple credit card delinquencies, or how much of it is coming from that
poorly strategized onboarding process.

➢ Analyze your pricing tiers


Pricing is fundamental; incorrectly setting price points and packaging products
in the wrong tiers can lead even a company with a great product to
underperform and a higher-than-necessary churn rate. Appropriate pricing
tiers are fundamental for conversions as well as minimal churn. Ensure that
your tiers are focused on buyer personas, not your own perceptions of your
features.

12
➢ Get metrics like ARPU, MRR, and ARR
Understanding churn is fundamental because of the effect churn can have on
your other key metrics. Retention analysis software shows you exactly how
the pieces all fit together to build user retention, reduce churn, and maintain
engagement.

The Good Type of Churn


There is, in fact, one good kind of churn, and that’s “negative churn.” Negative churn
is when your expansion revenue from all existing customers outweighs the revenue
lost from existing customers over the same period. Negative churn means that every
month, high yields will be coming in consistently from your customer base because
it’s a hugely powerful growth driver.

The only way to get your churn into those negative levels is with a strong, well-
thought-out approach to churn analysis. Understanding the directions that churn risk
comes from, and that each one represents an opportunity to improve your company
and strengthen your customer relationships, is the first step.

13
PART I: EXPLORING DATASET AND DATA DESCRIPTION

We will start our analysis with exploring our Dataset of Wiretel Telecommunication
Company. We will understand and interpret the results of Customer Churn in Wiretel
company by:

▪ Applying various data exploration and visualization techniques to describe the


customer churn from Wire Tel.
▪ Understanding the factors that explain customer churn.
▪ Suggesting reasons that why actually customers are churning from Wire Tel.
▪ Giving recommendations to the top management team of Wire Tel.
▪ Drafting a more focused retention strategy based on the analysis to the
company.

DATASET:

The dataset is extremely large and contains detailed information of all the parameters
which are extremely important for predictive churn analysis. The dataset consists of
3333 instances with 21 attributes, including the target variable, “CHURN”, an
indication of whether that customer churned (left the company) or not. The rich set
of attributes presented by the dataset helped in identifying customer churn more
effectively. It consists of both churned and not-churned customer types. The data is
input in CSV format and then visualized using various visual elements like charts,
graphs, etc. facilitating the better understanding of trends, outliers, and patterns in
the data.

14
TOOLS AND ML ALGORITHMS USED FOR ANALYSIS:

➢ Microsoft Excel:

➢ Data cleaning
Data cleaning is the process of fixing or removing incorrect, corrupted,
incorrectly formatted, duplicate, or incomplete data within a dataset. When
combining multiple data sources, there are many opportunities for data to
be duplicated or mis-labelled.

15
➢ Data preprocessing
Data preprocessing can refer to manipulation or dropping of data before it
is used in order to ensure or enhance performance, and is an important
step in the data mining process.

➢ Normalization & Standardization of Data


Normalization typically means rescales the values into a range of [0,1].
Standardization typically means rescales data to have a mean of 0 and a
standard deviation of 1 (unit variance).

➢ Correlation Matrix
A correlation matrix is simply a matrix which displays the correlation. It is
best used in variables that demonstrate a linear relationship between each
other. The matrix depicts the correlation between all the possible pairs of
values in a dataset.

➢ R Studio and Programming (Predictive Modelling):

➢ Linear Discriminant Analysis


LDA is a supervised dimensionality reduction technique. The goal is to
project a dataset onto a lower-dimensional space with good class-
separability in order avoid overfitting (“curse of dimensionality”) and
also reduce computational costs.

16
➢ Decision Tree

A decision tree is a decision support tool that uses a tree-like model of


decisions and their possible consequences, including chance event
outcomes, resource costs, and utility. It is one way to display an
algorithm that only contains conditional control statements.

➢ Neural Networks

A neural network is made up of densely connected processing nodes,


similar to neurons in the brain. Each node may be connected to different
nodes in multiple layers above and below it. These nodes move data
through the network in a feed-forward fashion, meaning the data moves
in only one direction. The node “fires” like a neuron when it passes
information to the next node.

➢ K-Nearest Neighbor (KNN)


KNN works by finding the distances between a query and all the
examples in the data, selecting the specified number examples (K)
closest to the query, then votes for the most frequent label (in the case
of classification) or averages the labels (in the case of regression).

➢ Clustering- K-Means

K-Means Clustering is an unsupervised learning algorithm that is used


to solve the clustering problems in machine learning or data science. It
is an algorithm which groups the unlabeled dataset into different
clusters. Here K defines the number of pre-defined clusters that need

17
to be created in the process, as if K=2, there will be two clusters, and
for K=3, there will be three clusters, and so on.

➢ Model-Wise Comparison
Checking which ML model has the best accuracy for better
interpretation and reliability of the results.

18
PART II: DATA VISUALIZATION

Depicting customer churn on above mentioned parameters and dataset:

➢ We have calculated the total percentage of customer left the company as per
the above data.

Churning Rate

14%

cust_churned
cust_retained

86%

Customer Churned 483


Customer Retained 2850
TOTAL 3333

From the above pie chart, we can infer that 483 customers churned from the
company which brings the churning percentage to 14% in Wiretel and we can see
that 2850 customers are retained which gives us company’s retention rate as 86%.
We can observe overall churning rate is not high enough. Customer that left are only
19
14% of the total customer population. This initially tells us that majority of customers
are satisfied by the Wiretel Services.

➢ Measuring Customer Churning state-wise:

States-Wise Churn Distribution Table

STATE WISE CHURN


120

100

80

60

40

20

0
AK AR CA CT DE GA IA IL KS LA MD MI MO MT ND NH NM NY OK PA SC TN UT VT WI WY

NO CHURN CHURN

20
Upon analyzing above graph, we can see company offers its services in 51 states.
The highest rate of customer churns are in state of CA 26.47% while highest rate of
non customer churns are in state of HI, 94.34%. Retention capacity of the HI state is
higher than those of other states. This could be due to the services, facilities,
schemes offered by Wiretel or due any other hidden factor. Further analysis will show
us the reasons for churning.

As per the given dataset, the company has served 3333 customers. So now we will
proceed to calculate the state-wise customer distribution i.e., customer
concentration in each state.

➢ Measuring State-Wise Customer Distribution

The Highest customer churning rate occurs in the states – NJ, TX, MD which means
the highest customers are leaving the company in these states. As per the graph
presented above, churning rate > 16%. And the lowest customer churning rate occurs

21
in the states – AK, HI, IA which means the customers are retained in these states. As
per the graph, churning rate < 4%.

➢ Customer churning in regard with Area Codes:


According to the dataset we have 3 Area Codes for 51 states. We will find out
the churning rate with respect to these arear codes

We have got three area codes in our dataset, 408, 415 and 510. From the above
graph we can clearly see that Area Code 510 has the highest churning rate while
Area Code 415 has the highest non- churning rate. Apart from that we can also infer
that the churning rate is almost equal for all the area codes. Therefore, we can say
that area wise customer churning is Homogeneous.

22
➢ Comparing Customer churning with Voice Mail Plans:

From the following graph, we can depict those customers who are not having the
voice mail plans are leaving the company.

3000

2500
No. of Customers

2000

1500
no

1000 yes

500

0
no yes
Voice_ mail plans

Above graph tells that Customers who have Voice Mail Plan have the highest non-
churning rate in the customer population. They constitute about 91.33% of the total
customers. We then find that 16.7% of those without the Voice Mail Plan are
churners, compared to 8.7% of customers who do have the Voice Mail Plan. Thus,
customers without the VoiceMail Plan are nearly twice as likely to churn as customers
with the plan. Here, 2411 customers does not have voice mail plans and hence they
are leaving the company.

23
➢ Comparing Customer churning with International Plans:

From the graph below, we can depict that the customers without international plans
are more likely to change their telecom company as compared to customers having
international plans.

3500

3000

2500
No. of customers

2000

no
1500
yes
1000

500

0
no yes
International Plan

Here, the graph shows that 3010 customers are without the international plans
whereas 323 are having the international plans. We can also observe that customers
with International Plan have a higher churning rate than those without it. Customers
who have International Plan and have churned constitute about 42.41% of the total
customer population while customers who don’t have International Plan and have
churned constitute remaining 57.59%. Thus from above analysis we can conclude
that customers with International Plan tend to churn higher than those without it.

24
➢ Comparing the Charges per minute made during the day, evening and night:

Charge/Min(in $)

0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
Day Eve. Night
Charge/Min 0.17 0.09 0.05

From the above graph, it can be seen that during the day the charge per minute by
the company is highest and charge per minute made in night is at the lowest level.

➢ Total calls made by the customers:

Total calls

28137
27962

27775

DAY CALLS EVE. CALLS NIGHT CALLS

From the above graph, we can infer that 27,962 calls were made in the day time,
28,137 were made in the evening and 27,775 calls were made at night. So, we can
infer that most preferred time for calls is Evening followed by, Day Calls.
25
PART III: DATA PROCESSING AND LINEAR
DISCRIMINANT ANALYSIS

DATA PRE-PROCESSING:

Before beginning with the actual analysis, we will first check our dataset for any
missing values using R-Programming. The following code will show us that if there
is any NA value in our dataset and this is an important step as it helps in getting
better and error-free results for further analysis.

As it can be seen from the above output there are 0 missing values in the dataset.

DESCRIPTIVE STATISTICS:

For further analysis, we will apply Descriptive Statistics on our dataset using
Microsoft Excel. Descriptive statistics are used to describe or summarize the
characteristics of a sample or data set, such as a variable's mean, standard deviation,
or frequency. Inferential statistics can help us understand the collective properties of
the elements of a data sample.

26
Descriptive statistics are broken down into measures of central tendency and
measures of Dispersion. Measures of central tendency include the mean, median,
and mode, while measures of variability include Standard Deviation, Variance,
Range, and InterQuartile Range. Also, Descriptive statistics are brief descriptive
coefficients that summarize a given data set, which can be either a representation of
the entire population or a sample of a population.

For descriptive statistics we will first convert all the categorical variables in
continuous values, i.e.,
• Churn: True/False – 1/0
• VoiceMail Plan: Yes/No – 1/0
• International Plan: Yes/No – 1/0
After converting the values to binary values, our dataset will look like the following.

27
DATA SUMMARIZATION:
Summary statistics summarize and provide information about your sample data. It
tells you something about the values in your data set. This includes where
the mean lies and whether your data is skewed.
We have calculated descriptive statistics using Data Analysis tab in Microsoft Excel.

28
29
The above tables show the measures of central tendency as well as measures of
variability in the dataset.

➢ The minimum account length is one month, the maximum is 243 months, and
the mean and median are about the same, at around 101 months, which is an
indication of symmetry. Several variables show this symmetry, including all
the minutes, charge, and calls field.

30
➢ Mean in case of International Plan is 0.09691 which is close value to 0.1
indicating that most of the customers do not have International Plan. Same
also applies to Voice Mail Plan, mean value is close to 0.

➢ There is no symmetry in voice mail messages and customer service calls. The
median for voice mail messages is zero, indicating that at least half of all
customers had no voice mail messages. This results, of course, from fewer
than half of the customers selecting the Voice Mail Plan.

➢ The mean of customer service calls (1.563) is greater than the median (1.0),
indicating some right-skewness, as also indicated by the maximum number of
customer service calls being nine.

LINEAR DISCRIMINANT ANALYSIS (LDA)


Linear Discriminant Analysis, or LDA, is a linear machine learning algorithm used for
multi-class classification. LDA is used for compressing supervised data. When we
have a large set of features (classes), and our data is normally distributed and the
features are not correlated with each other then we can use LDA to reduce the
number of dimensions. LDA is a generalized version of Fisher’s linear discriminant.
It seeks to best separate (or discriminate) the samples in the training dataset by their
class value. Specifically, the model seeks to find a linear combination of input
variables that achieves the maximum separation for samples between classes (class
centroids or means) and the minimum separation of samples within each class. We
have used the LDA algorithm to find out the features which are useful for further
analysis in context with our target variable “Churn”.

31
Step 1:
Installing the Packages and calling the required functions using the ‘Library( )’
command.

Step 2:
Extracting the data.

Step 3:
Plotting the Pairs-Panels Matrix.

32
33
Step 4:
Data Partition

34
Step 5:
Using LDA function and printing the result.

Step 6:
Getting the attributes.

35
Step 7:
Plotting the Histogram

36
Step 8:
Creating the Confusion Matrix and checking the accuracy of Training and Testing
Data.

Interpretation:
Our Training Data has the accuracy of 85.51% while our Testing Data has accuracy
of 86.58%. This means we can safely say that the model is correctly formulated and
predicted. The Testing Data showed higher result than Training Data which can be
seen as a sign of a successful model. Also the accuracy rate classifier reached above
85% which is pretty good, considering we have a large dataset with only two classes.

37
PART IV: FACTORS THAT EXPLAIN CUSTOMER CHURN

For determining the factors that explain customer churn we will follow the below 4
steps:

1. Normalization Of Data
2. Finding out the further Correlation between variables by plotting the
Correlation matrix
For the purpose dimension reduction of our normalized dataset.
3. Machine Learning/Deep Learning Algorithms such as Decision Tree analysis,
Artificial Neural Networks, KNN, and Clustering
4. Model-Wise Comparison

As per the process we will first proceed with the Normalization Of Data, which was
done in MS-Excel using the below formula:

Normalization is used to scale the data of an attribute so that it falls in a smaller


range, such as -1.0 to 1.0 or 0.0 to 1.0. It is generally useful for classification
algorithms. Normalization is generally required when we are dealing with attributes
on a different scale, otherwise, it may lead to a dilution in effectiveness of an
important equally important attribute(on lower scale) because of other attribute
having values on larger scale. In simple words, when multiple attributes are there

38
but attributes have values on different scales, this may lead to poor data models
while performing data mining operations. So they are normalized to bring all the
attributes on the same scale.

After normalizing the dataset, we will have following values:

Correlation Matrix:

A correlation matrix is a table showing correlation coefficients between variables.


Each cell in the table shows the correlation between two variables. A correlation
matrix is used to summarize data, as an input into a more advanced analysis, and as
a diagnostic for advanced analyses.

Correlation matrix, is also used to investigate the dependence between multiple


variables at the same time. The result is a table containing the correlation
coefficients between each variable and the others.
There are different methods for correlation analysis : Pearson parametric correlation
test, Spearman and Kendall rank-based correlation analysis.

39
Excel

RStudio

Correlation Matrix in R using Cor function.

40
Plotting the Matrix using ‘corrplot( )’ function.

Once we have found out the correlation we will proceed to dimension reduction to
make data more organized and relevant. This is basically reducing the
features/variables to make data more comprehensive and readable. For Dimension
Reduction we will check the coefficient of correlation of the variables or the features
with correlation 1 (perfect linear correlation) would be selected. Then among the two
features with linear collinearity, one of them would be dropped. A further benefit of
doing so is that the dimensionality of the solution space is reduced, and the
algorithms may work more efficiently to find the optimal solution. Since in the data
set, day minutes and day charge, night minutes and night charge, evening minutes

41
and evening charge and international minutes and international charge showed
perfect correlation I decided to drop the charge column of each type. After dimension
reduction we will be left with 14 variables and 1 target variable. The data set will look
something like this:

Since our dataset is now reduced and normalized we can proceed with the machine-
learning algorithms.

DECISION TREE

A decision tree is a decision support tool that uses a tree-like model of decisions and
their possible consequences, including chance event outcomes, resource costs,
and utility. It is one way to display an algorithm that only contains conditional control
statements. A decision tree is a flowchart-like structure in which each internal node
represents a "test" on an attribute, each branch represents the outcome of the test,
and each leaf node represents a class label (decision taken after computing all
attributes). The paths from root to leaf represent classification rules.

42
A decision tree consists of three types of nodes:[1]

1. Decision nodes – typically represented by squares


2. Chance nodes – typically represented by circles
3. End nodes – typically represented by triangles
Step 1:

Data Partition.

Step 2:

Plotting the Decision Tree.

Step 3:
43
Creating Confusion Matrix and checking overall accuracy of Decision Tree

Interpretation:

The confusion matrix displays the following information:

• true positives (TP): These are cases in which we predicted yes (they
churned), and they did churn.
• true negatives (TN): We predicted no, and they didn’t churn.
• false positives (FP): We predicted yes, but they didn’t actually churn. (Also
known as a “Type I error.”)
• false negatives (FN): We predicted no, but they actually churned. (Also
known as a “Type II error.”)

The diagonal entries give our correct predictions, with the upper left being TN and
the lower right being TP. The upper right gives the FN while the lower left gives the
FP. From this confusion matrix, we can see that the model performs well at
predicting non-churning customers (851 correct vs. 12 incorrect) as well as at
predicting churning customers (106 correct vs. 30 incorrect).

44
The decision tree model is mostly accurate, correctly predicting the churn status of
customers in the test subset 95.7% of the time.

NEURAL NETWORK

A neural network evaluates price data and unearths opportunities for making trade
decisions based on the data analysis. The networks can distinguish subtle nonlinear
interdependencies and patterns other methods of technical analysis cannot. There
will always be data sets and task classes that a better analyzed by using previously
developed algorithms. It is not so much the algorithm that matters; it is the well-
prepared input data on the targeted indicator that ultimately determines the level of
success of a neural network.

There are three main components: an input later, a processing layer, and an output
layer. The inputs may be weighted based on various criteria. Within the processing
layer, which is hidden from view, there are nodes and connections between these
nodes, meant to be analogous to the neurons and synapses in an animal brain.

Step 1:

Loading the required library and getting the structure of the data using ‘str()’ function.

45
Step 2:

Applying the Neural Network Algorithm and plotting the graph.

Step 3:

Displaying the ANN (Artificial Neural Network) weights.

46
Step 4:

Creating the Result Matrix.

Step 5:

Data Partition and Predicting the result.

47
Step 6:

Creating the Confusion Matrix and checking the accuracy of Training Data.

Step 7:

Testing the resulting output.

Step 8:

Creating the Confusion Matrix to check the accuracy of Testing Data.

48
Interpretation:

As per our results Training Data has an accuracy of 68.2% while Testing Data has an
accuracy of 68.3%. We can see overall, the neural network model is not performing
well and giving us non-accurate results by a big margin on both the train and test
sets. The model is suitable for further analysis.

K-NEAREST NEIGHBOR

The K-nearest neighbours (KNN) algorithm is a simple, supervised machine learning


algorithm that can be used to solve both classification and regression problems. It’s
easy to implement and understand, but has a major drawback of becoming
significantly slows as the size of that data in use grows. KNN works by finding the
distances between a query and all the examples in the data, selecting the specified
number examples (K) closest to the query, then votes for the most frequent label
(in the case of classification) or averages the labels (in the case of regression).

49
Step 1:

Getting the data.

Step 2:

Data Partition.

Step 3:

Checking the dimensions of the data and looking for missing values.

Step 4:

Getting the summary of the data.

50
Step 5:

Applying KNN and plotting the graph.

51
Step 6:

Checking the Predictive Capability.

52
Step 7:

Confusion Matrix and and checking accuracy of KNN.

Interpretation:

So, from the output, we can see that our model predicts the outcome with an
accuracy of 90.39% which is really good considering we worked with a large data
set. A point to remember is that the more data (optimal data) you feed the machine,
the more efficient the model will be. K is 5 which is more than when k was 1,2,3
and 4. Same accuracy when K was 15 which means now increasing k values
doesn’t affect the accuracy.

53
CLUSTERING

Clustering or cluster analysis is a machine learning technique, which groups the


unlabelled dataset. It can be defined as "A way of grouping the data points into
different clusters, consisting of similar data points. The objects with the possible
similarities remain in a group that has less or no similarities with another group." It
does it by finding some similar patterns in the unlabelled dataset such as shape, size,
color, behavior, etc., and divides them as per the presence and absence of those
similar patterns. It is an unsupervised learning method, hence no supervision is
provided to the algorithm, and it deals with the unlabeled dataset. After applying this
clustering technique, each cluster or group is provided with a cluster-ID. ML system
can use this id to simplify the processing of large and complex datasets.

Step 1:

Loading the required Libraries.

Step 2:

Plotting the Correlation Matrix.

54
Step 3:

Applying Silhouette Score and Plotting the Elbow Curve.

55
Step 4:

Applying K-Means to see Optimal number of Clusters.

56
Step 5:

Adding Cluster Classification to our Data

57
Interpretation:

As we can see in above table, our data is classified into two clusters 1 and 2.

58
MODEL-WISE COMPARISON

Decision Tree was found to be giving really high accuracy, 95%, which could indicate
the case of overfitting of data that’s why we will not consider it for further analysis.
Among the remaining models KNN was giving the highest accuracy at 90%.
Therefore, the KNN model will be considered for further analysis.

59
PART V: WHY CUSTOMERS ARE CHURNING FROM
WIRETEL?

Based on the above findings we can say that the customers are leaving from Wiretel
because of the following reasons and assumption:

REASONS ASSUMPTIONS
International Plan They don’t want to opt this plan.
Voice Mail Plan They want more benefits to be included
Day Charges They want it to be decreased.
Evening Charges They want it to be decreased.
Night Charges They want it to be decreased.
International Charges They want it to be decreased.
International Calls They may want better connection and
network plans or may be improved data
packs for smooth communication.
Customer Service Based on the analysis we can say that
excessive customer service calls aren’t
desired by customers, hence company
should reduce them, in order to retain
customers.

60
PART VI: RECOMMENDATION BE TO THE TOP
MANAGEMENT TEAM OF WIRETEL

• Customers with the International Plan tend to churn more frequently. So


Wiretel should stop offering International Plans and bring out a fresh plan to
substitute it.

• Customers with the Voice Mail Plan tend to churn less frequently. Since
churning rate is low in case of Voice Mail Plan, the company could offer more
of these to the customers, because they seem to be satisfied with the benefits
it contains.

• Customers with four or more customer service calls churn more than four
times as often as do the other customers. Therefore, service calls should be
reduced to retain the customers.

• Customers with high day minutes and evening minutes tend to churn at a
higher rate than do the other customers. To overcome that Day Charges and
Evening Charges should be increased, as it will ultimately reduce the day
minutes and evening minutes.

• There is no obvious association of churn with the variables day calls, evening
calls, night calls, international calls, night minutes, international minutes,
account length, or voice mail messages. So company should start focusing on

61
the significant variables as analyzed above, that are related to and influence
Churning. This will reduce the customer churn.

• The model tells us the significant variables and their corresponding impact on
churn. With this model, the company can now have great insight into which
customers may discontinue their service. It can offer better deals and
discounts to customers who are likely to stop using the service.

• States with high churn rates should be given special consideration as it is


important to understand what is causing the churn in those customers. It could
be that they are not satisfied with the service quality. Considering the high
churn rate in CA, Wiretel should examine the situation carefully.

• Company can offer good deals during the first 1-2 months of service and can
increase prices afterward. As the data shows the customers discontinuing the
service have a mean of around 3.5 months.

• Wiretel has collected most of its revenue from MN, NY, and WV since its
foundation; therefore, the company should also concentrate on these states
for more growth.

• Only a very small number of customers are currently subscribed to the


company's International and Vmail plans. Further investigations needs to be
performed to understand what customers don’t like about the overall service.

62
PART VII: DRAFTING A MORE FOCUSED RETENTION
STRATEGY FOR WIRETEL

A more focused retention strategy will include the objectives, reduction of churning
rate and to maximization of revenue. In order to carry out the business retention
strategy, a careful mapping of customer activities needs to be carried out.

Analysis of their trends, buying behavior and usage pattern should be done in order
to segregate the customers according to our objectives and goals. As it is not feasible
to retain all the churners, it is necessary to decide which customer churners have
the potential of turning into the non-churners and remaining customers who can’t be
turned into non-churners can be excluded.

Our analysis lead us to the finding that customers with the International Plan tend to
churn more frequently so International Plans should be replaced with other plan or
should be upgraded to something that customers find more lucrative. That includes
introducing new features and add on benefits. But where International Plan fails,
Voice Mail Plan succeeds as customers with the Voice Mail Plan tend to churn less
frequently. Since churning rate is low in case of Voice Mail Plan, the company could
offer more of these to the customers, because they seem to be satisfied with the
benefits it contains.

Apart from that Company could also offer other benefits like improved data services,
better communication models, top ups and add on benefits, various occasional
schemes and discounts, revised charges, and other facilities. To conclude, Company

63
should carefully examine the reasons for customer churn and overcome them or try
to build upon the reasons which are play part in customer retention. Wiretel should
stop wasting it resources on less significant factors and try to invest in factors which
could actually give high ROI.

➢ Reasonable Tariff Plan(s) :


As we can see from the data analysis , customers are leaving wiretel due to
different calls charges. For example the majority of customers are leaving due
to Day Charges/tariff plans. If competitive and reasonable tariff plans are
imposed then wirtel can saves a lot and can increase the retention.

➢ Innovate through Technology :

The reason of customer churn from the telecome operators are sometimes,
due to bad network. And Telecom organization faces problems in pinpointing
the location of network faults so through “IoT ans Sensors” , problem of
network faults can be solved because it impacts the time value of customers
by eliminating the effects of length of services.

➢ Customer Feedback :

Customer Churn does not usually strike out of the blue. More likely, a desire
to leave has built up over time as customers grow increasingly disgruntled
with their experience. From the analysis of this case, I think top management
must encourage customer care desk to listen to customers first why are the
unhappy about services and then tackle the issue because it will encourage
the customers to stick around.

64
➢ Implementation of Chatbot to provide more human interactions:
As the data of customers from USA and customers are tech heavy so providing
chatbot would provide feel of human because customers would like more
personal interactions and IVR does not provide this. I assume that wiretel
might be using IVR and they should replace it with intelligent agents like
chatbot.

➢ Implementation of Relevant Loyalty Programs:


From this case analysis I found that the customers are churning due to
improper loyalty programs. The insights from the customer data shows a clear
trend of customers leaving due to higher day tariff plans so top management
must offer tailored loyalty plans based on usage.

From the State Vs Account length visuals/Plot , we can see that trends shows that
there are states where more loyal customers , top management can mark those

65
states as high performing states and customized loyalty programs should be rolled
out to increase retention.

The company should keep an eye on the pricing of the competitors as well as the
deals they are offering, which helps to will switch to your competitors. anticipate
whether or not the customers.

The top management should carefully examine the results of the analysis and
identify the patterns, errors and metrics for better understanding of their
customers. Customer retention is the most important strategy for any company, as
a result the focus should be more on the retention process rather than expansion
or diversification of the business. Unnecessary schemes, policies, discounts, offers
should be ruled with immediate effect as they are not preferred by the customers.
Only those services should be offered to customers which offer them value as well
as create wealth for the company. Customers are the only source to earn profits in
the long term, hence if company ignores this aspect of the business, it may have
promising returns initially but in long term the current strategy will not help the
company at all, since losing customers means losing revenue.
Focus should be more on Customer’s preferences and buying behaviour. Also the
purchasing patterns these will help to the company to concentrate it’s operations in
those areas/dimensions only, and will further avoid wastage of time, money,
resources and last but not the least customer potential.
If company want’s to survive in long term, it should definitely shift it focus on
customer preferences and needs.

66
SUSTAINABLE DEVELOPMENT GOALS

➢ Telecommunications infrastructure is an enabler for Smart Cities


The transformational possibilities 5G can bring really demonstrates the impact
the telecoms industry has in helping achieve the SDG’s; from increasing
efficiency and productivity at traffic junctions, to providing remote care, to
optimising manufacturing processes, to increasing emergency response times,
to providing socio-economically poorer communities a chance to be digitally
included.

➢ Smarter Cities will deliver a healthier and more sustainable world


5G will be critical because it will provide super fast broadband, ultra-reliable
low latency communication, massive machine-type communications, high
reliability/availability and efficient energy usage.

➢ Decent Work & Economic Growth:


A 10% increase in mobile broadband adoption means a 0.8% increase in a
nation’s gross domestic product, according to research conducted by Ericsson
with London’s Imperial College.

67
➢ Industry, Innovation & Infrastructure
Half the world’s population already has access to the internet — the next 50%
will be connected by broadband. And the 5.9 billion mobile broadband
subscriptions worldwide today will surpass 8 billion by 2024. (Many people
have multiple subscriptions.) This will enable social and economic inclusion
and help meet societal needs in education, health, e-government and
entrepreneurship.

➢ Responsible Consumption & Production


5G technology will enable factories to be more flexible, safer, energy efficient,
and cost-effective.

➢ Climate Action
Digitalisation of a range of services and industries, which will be aided by 5G,
can help reduce greenhouse and gas emissions by up to 15% by 2030.

The telecoms industry plays a critical role in achieving the Sustainable Development
Goals and is a key enabler for many solutions to be successful. The more widespread
wireless infrastructure and mobile coverage is, the more likely it is that entrepreneurs
will be able to innovate and create applications that can impact society in a positive
way. Investing in digital infrastructure should be a global priority; telecoms should join
other utility industries such as energy & water as an essential infrastructure
requirement.

68
BIBLIOGRAPHY

1)

https://indjst.org/download-article.php?Article_Unique_Id=INDJST2058&Full_Text_Pdf_Download=True

2)

https://towardsdatascience.com/customer-churn-in-telecom-segment-5e49356f39e5

3)

http://www.ijfcc.org/vol8/550-D0015.pdf

4)

http://ce.sharif.edu/courses/85-86/1/ce925/assignments/files/assignDir4/Churn.pdf

5)

https://books.google.co.in/books?id=qbfEBgAAQBAJ&pg=PA80&lpg=PA80&dq=Day+Charge+%3D+0.000
613+%2B+0.170+Day+Mins&source=bl&ots=VjDhOxpIJl&sig=ACfU3U0MizmjmUoxiRtMRpykLrO4n1Y4Dg
&hl=en&sa=X&ved=2ahUKEwju7-DX4bv1AhVR_XMBHYpjAsAQ6AF6BAgCEAM#v=onepage&q&f=false

69

You might also like