Data Mining to Classify Telco Churners

Data Mining To
Classify Telco
Churners
Mohit Mhapuskar
Guided by Prof. Mahmoud Daneshmand

Introduction
● In the modern world, almost all the civilized population subscribe to a
telecommunications provider for their cellular connections, internet
connections, landline, television connection etc.
● However, many customers often switch or unsubscribe (churn) from their
telecom providers for a variety of reasons. These could range from
unsatisfactory service, better pricing from competitors, customers moving
to different cities etc.
● Therefore, telecom companies are interested in analyzing the patterns for
customers who churn from their services and use the resultant analysis to
determine in the future which customers are more likely to unsubscribe
from their services. One such company is Telco Systems. Telco Systems is
interested in identifying the precise patterns for their churning customers
and have provided the customer data for this project.
2

Standard Industry Process : CRISP -DM
3
1 2 3
4 5 6

Business Understanding
● What is the profound question?
● Whether a subscriber will churn from the telecom service or not?
● What is the project goal?
● Use data mining techniques such as classification algorithms to determine
whether customers will churn or not churn and generate the classification
rules accordingly.
● What is the business implication?
● The generated rules will allow Telco systems to understand the factors as to
why customers may churn from their service. It will also enable them to
modify their business practices accordingly, to reduce the number of
churning customers in the future.
5

Data Understanding
● Data Source
● The data provided by Telco Systems contains the customer churn records in
the span of one month. Dataset was obtained from www.Kaggle.com.
● Dataset Details
● The raw dataset comprises 7043 records of customers along with 21
attributes. 20 of these attributes are independent attributes.
● Target Attribute
● The target attribute is the one that classifies the subscribers as having
churned from the service or not.
7

Data Understanding
Dataset Sample
8

Data Understanding
Understanding the attributes
1. CustomerID: Company assigned ID (Unique Primary Identifier).
2. Gender: Customer gender (Female, Male).
3. SeniorCitizen: Whether the customer is senior citizen (1, 0).
4. Partner: Whether the customer has a partner (Yes, No).
5. Dependents: Whether the customer has dependents (Yes, No).
6. Tenure: No. of months customer has stayed with the company (Continuous
Variable).
7. PhoneService: Whether the customer has phone service (Yes, No).
8. MultipleLines: Whether the customer has multiple lines (Yes, No, No Internet
Service).
9. InternetService: Type of customer’s Internet service (DSL, Fiber Optic, No).
10. OnlineSecurity: Whether the customer has internet service security (Yes, No,
No Internet Service).
11. OnlineBackup: Whether the customer has online backup plan (Yes, No, No
Internet Service).
9

Data Understanding
12. DeviceProtection: Whether the customer has internet device protection (Yes,
No, No Internet Service).
13. TechSupport: Whether the customer has tech support (Yes, No, No Internet
Service).
14. StreamingTV: Whether the customer has streaming TV (Yes, No, No Internet
Service).
15. StreamingMovies: Whether the customer has streaming movies (Yes, No, No
Internet Service).
16. Contract: The contract term of the customer (Month-to-Month, One Year, Two
Year).
17. PaperlessBilling: Whether customer has paperless billing (Yes, No).
18. PaymentMethod: Customer’s payment method (Electronic check, Mailed check,
Bank transfer (automatic), Credit card (automatic) ).
19. MonthlyCharges: Customer’s monthly charges (Continuous Variable).
20. TotalCharges: Customer’s total charges (Continuous Variable).
21. Churn: Whether customer churned or not (Yes, No) (Target Variable).
10

Data Preparation
12
Select Appropriate Variables for Analysis
Perform Attribute Transformations If Needed
Handle Missing Values & Outliers If Needed1
2
3
In this phase the dataset will be cleaned and transformed to prepare
it accordingly for analysis. Data preparation may involve any of the
3 following operations.

Data Preparation
Handling Missing Values
The dataset has missing values for 11 records in the attribute TotalCharges as
shown below. Since there are only 11 such records out of a total of 7043, we can
safely eliminate these records from our dataset without impacting the overall
accuracy of our model.
13

Data Preparation
Identifying Outliers
The dataset was checked for outliers. There were no outliers detected in the dataset.
Attribute Selection
All the independent variables were deemed important in classification. All the
attributes contribute toward the final target variable i.e. whether a customer churns
or not.
The final list of variables is therefore: gender, SeniorCitizen, Partner, Dependents,
Tenure, PhoneService, MultipleLines, InternetService, OnlineSecurity, OnlineBackup,
DeviceProtection, TechSupport, StreamingTV, StreamingMovies, Contract,
PaperlessBilling, PaymentMethod, MonthlyCharges, TotalCharges.
14

Data Preparation
Data Division
The final transformed dataset consisting of 7033 records will be split into 2 subsets.
15
Training Dataset
The Training Dataset will be
used to generate our model.
80% of the data i.e. about 5626
records will be used for training.
Test Dataset
The model generated using the
training dataset will be applied
to the test dataset to evaluate
and test the accuracy of the
model. 20% of the data i.e. the
remaining 1407 records will be
used for testing.

Modeling
● Now that we have prepared the data for analysis, the next phase begins
with the selection of an appropriate data modeling technique. We will use
one of the Machine Learning algorithms taught in this course for our project.
● Since the main goal of this project is to classify the telecom service
subscribers as Churners and Non-Churners, Recursive Partitioning is ideal
since it recursively splits the data at each node into subsets resulting in a
decision tree.
● As the target variable is a binary categorical variable (Churn – Yes or No),
Classification trees will be most appropriate for this dataset.
● We conclude that the Classification and Regression Trees or CART algorithm
is the most optimal solution for this project.
17

Modeling
18
Modeling Software
KNIME is a free and open-source data analytics, reporting and integration platform.
KNIME integrates various components for machine learning and data mining through
its modular data pipelining concept.
We will use the KNIME software and its in-built modules to import the dataset, split
the dataset, generate our CART decision tree along with its decision rules and test
the accuracy of the model on the test dataset using its in-built scorer.
Software obtained from https://www.knime.com/knime-software/

Modeling
19
Process Workflow
1. File Reader Imports
Dataset
2. Partitioner splits into
80% Training and 20%
Test Data.
3. Decision Tree Learner
generates CART Tree.
4. Decision Tree Predictor
applies model to test
dataset.
5. Ruleset is generated
and written.
6. The Scorer evaluates
accuracy of the model.

Modeling
20
CART Decision Tree

21
Modeling
21
Generated Decision Rules
Antecedent Consequent
$StreamingMovies$ = No AND $Contract$ = One year No
$TotalCharges$ <= 863.55 AND $TotalCharges$ <= 1166.35 AND
$MonthlyCharges$ <= 99.85 AND $StreamingMovies$ = Yes AND $Contract$ = One
year
No
$TotalCharges$ > 863.55 AND $TotalCharges$ <= 1166.35 AND $MonthlyCharges$
<= 99.85 AND $StreamingMovies$ = Yes AND $Contract$ = One year
Yes
$PaymentMethod$ = Mailed check AND $DeviceProtection$ = Yes AND
$TotalCharges$ <= 6552.75 AND $MonthlyCharges$ > 99.85 AND
$StreamingMovies$ = Yes AND $Contract$ = One year
No
$PaymentMethod$ = Bank transfer (automatic) AND $DeviceProtection$ = Yes AND
No
$OnlineSecurity$ = Yes AND $PaymentMethod$ = Electronic check AND
$DeviceProtection$ = Yes AND $TotalCharges$ <= 6552.75 AND $MonthlyCharges$
> 99.85 AND $StreamingMovies$ = Yes AND $Contract$ = One year
No

22
Modeling
22
$TechSupport$ = No AND $OnlineSecurity$ = No AND $PaymentMethod$ =
Electronic check AND $DeviceProtection$ = Yes AND $TotalCharges$ <= 6552.75
AND $MonthlyCharges$ > 99.85 AND $StreamingMovies$ = Yes AND $Contract$ =
One year
No
$TechSupport$ = Yes AND $OnlineSecurity$ = No AND $PaymentMethod$ =
Electronic check AND $DeviceProtection$ = Yes AND $TotalCharges$ <= 6552.75
AND $MonthlyCharges$ > 99.85 AND $StreamingMovies$ = Yes AND $Contract$ =
One year
Yes
$Partner$ = No AND $OnlineBackup$ = Yes AND $DeviceProtection$ = No AND
No
$Partner$ = Yes AND $OnlineBackup$ = Yes AND $DeviceProtection$ = No AND
Yes
$MonthlyCharges$ <= 29.497 AND $PaperlessBilling$ = No AND $SeniorCitizen$ <=
0.5 AND $TotalCharges$ <= 81.7 AND $TotalCharges$ <= 310.9 AND
$InternetService$ = DSL AND $Contract$ = Month-to-month
No

23
Modeling
23
$OnlineSecurity$ = Yes AND $MonthlyCharges$ > 29.497 AND $PaperlessBilling$ =
No AND $SeniorCitizen$ <= 0.5 AND $TotalCharges$ <= 81.7 AND $TotalCharges$
<= 310.9 AND $InternetService$ = DSL AND $Contract$ = Month-to-month
No
$MultipleLines$ = No AND $PaymentMethod$ = Mailed check AND
$MonthlyCharges$ <= 55.425 AND $TotalCharges$ > 81.7 AND $TotalCharges$ <=
310.9 AND $InternetService$ = DSL AND $Contract$ = Month-to-month
No
$MultipleLines$ = Yes AND $PaymentMethod$ = Mailed check AND
$MonthlyCharges$ <= 55.425 AND $TotalCharges$ > 81.7 AND $TotalCharges$ <=
310.9 AND $InternetService$ = DSL AND $Contract$ = Month-to-month
Yes
$TechSupport$ = No AND $MonthlyCharges$ > 50.85 AND $PaymentMethod$ =
Mailed check AND $OnlineBackup$ = No AND $TotalCharges$ <= 1175.1 AND
$TotalCharges$ > 310.9 AND $InternetService$ = DSL AND $Contract$ = Month-to-
month
No
$tenure$ <= 8.5 AND $TotalCharges$ <= 897.45 AND $PaymentMethod$ = Electronic
check AND $OnlineBackup$ = No AND $TotalCharges$ <= 1175.1 AND
$TotalCharges$ > 310.9 AND $InternetService$ = DSL AND $Contract$ = Month-to-
month
No

24
Modeling
24
$OnlineBackup$ = Yes AND $TotalCharges$ <= 1175.1 AND $TotalCharges$ > 310.9
AND $InternetService$ = DSL AND $Contract$ = Month-to-month
No
$TotalCharges$ <= 120.0 AND $tenure$ <= 13.5 AND $InternetService$ = Fiber optic
AND $Contract$ = Month-to-month
Yes
$PaperlessBilling$ = No AND $MultipleLines$ = No AND $MonthlyCharges$ <=
75.675 AND $TotalCharges$ > 120.0 AND $tenure$ <= 13.5 AND $InternetService$ =
Fiber optic AND $Contract$ = Month-to-month
No
$PaymentMethod$ = Electronic check AND $TotalCharges$ <= 36.325 AND
$InternetService$ = No AND $Contract$ = Month-to-month
No
$TotalCharges$ > 6786.75 AND $OnlineSecurity$ = No AND $tenure$ > 54.5 AND
$tenure$ > 13.5 AND $InternetService$ = Fiber optic AND $Contract$ = Month-to-
month
Yes
$OnlineSecurity$ = Yes AND $tenure$ > 54.5 AND $tenure$ > 13.5 AND
$InternetService$ = Fiber optic AND $Contract$ = Month-to-month No

Evaluation
● Using Cross-Validation to test the accuracy of the model, we apply our
generated model on the test data to determine the accuracy of the model.
● From our results, we can see that about 78% of the customer records were
correctly classified.
26

Deployment
● The Deployment phase is the final phase in the CRISP-DM process. Here
we will deploy our generated model.
● In this phase of our project, the resultant analysis from our model is
compiled in the form of simple-to-read reports and shared with the
executive board of Telco Systems.
● The results can help Telco Systems understand the factors as to why
customers may unsubscribe from their services.
● They can use that data to modify their existing services and business
practices or launch new company initiatives to better entice customers to
stay with their service.
28

References
● Dataset obtained from www.kaggle.com/blastchar/telco-customer-churn
● Software obtained from www.knime.com/knime-software/knime-analytics-platform
30

Data Mining to Classify Telco Churners

More Related Content

What's hot

Similar to Data Mining to Classify Telco Churners

Recently uploaded

Data Mining to Classify Telco Churners