PM-GA-SS.pdf - Predictive Modeling Assignment Prediction of...

Predictive Modeling Assignment
– Prediction of Customer Churn
Model Solution
1

Table of Contents
1 Phase I – Discovery .............................................................................. 3
1.1 The Business Domain ...................................................................... 3
1.2 The Problem in hand ....................................................................... 4
1.2.1 Cell Phone Data Attributes .......................................................... 4
1.3 The Initial Hypothesis ...................................................................... 4
2 Phase II – Data Preparation ................................................................... 5
2.1 Data Exploration and Visualization .................................................... 5
2.1.1 Initial Data Exploration .............................................................. 5
2.2 Data Preparation – Training and Testing Dataset ................................. 6
3 Phase III – Model Planning and Building .................................................. 7
3.1 Model Selection and initial build ........................................................ 7
3.1.1 Model 1 with all variables: .......................................................... 7
3.2 Model 1 Significance Verification ....................................................... 8
3.2.1 Overall Significance of the Model – Log Likelihood Ratio Test .......... 8
3.2.2 Model Robustness – McFadden’s pseudo -R Squared Test ................ 9
3.2.3 Test for Individual Coefficients .................................................... 9
3.2.4 Odds Explanatory Power ........................................................... 10
3.2.5 Classification Table .................................................................. 11
3.2.6 ROC Plot ................................................................................ 13
3.3 Model Refining .............................................................................. 14
3.3.1 Variable selection using Step Function ....................................... 14
3.3.2 Model 2: Using step() function .................................................. 16
3.4 Model 2: Significance Verification .................................................... 17
3.4.1 Overall Significance of the Model – The Log Likelihood Test .......... 17
3.4.2 Model Robustness – McFadden’s pseudo -R Squared Test .............. 17
3.4.3 Test for Individual Coefficients .................................................. 17
3.4.4 Odds Explanatory Power ........................................................... 18
3.4.5 Classification Table .................................................................. 19
3.4.6 ROC Plot ................................................................................ 20
3.5 Model Refining – Uncovering interactions ......................................... 21
4 Phase IV – Communicating Results ....................................................... 22
4.1 Conclusion on final Model ............................................................... 22
4.2 Final interpretation ........................................................................ 22

3 | P a g e
1 Phase I – Discovery
1.1 The Business Domain
Customer attrition is an important issue for any industry. It is especially
important in mature industries where the initial period of exponential
growth has been left behind. Not surprisingly, attrition (or, to look on the
bright side, retention) is a major application of data mining.
One of the first challenges in modeling attrition is deciding what it is and
recognizing when it has occurred. This is harder in some industries than in
others. At one extreme are businesses that deal in anonymous cash
transactions.
When a once-loyal customer deserts his regular coffee bar for another
down the block, the barista who knew the customer's order by heart may
notice, but the fact will not be recorded in any corporate database. Even in
cases where the customer is identified by name, telling the difference
between a customer who has churned and one who just hasn't been around
for a while may be hard. If a loyal Ford customer who buys a new F150
pickup every five years hasn't bought one for six years, has the customer
defected to another brand?
Attrition is a bit easier to spot when a monthly billing relationship exists, as
with credit cards. Even there, attrition might be silent. A customer may
stop using the credit card, but not cancel it. Attrition is easiest to define in
subscription-based businesses, and partly for that reason, attrition
modeling is most popular in these businesses. Long-distance companies,
mobile phone service providers, insurance companies, cable companies,
financial services companies, Internet service providers, newspapers,
magazines, and some retailers all share a subscription model where