Analysing Churn for a mobile service provider

Project Details


2012 (Nov)


Farah Sarfraz, Niharika Vempati, Pallavi Sabharwal, Shilpa Murthy





One of the most critical factors in Customer Relationship Management that can make or break
a company’s long-term profitability is churn. If a company can predict whether a customer is
likely to churn, it can take a more targeted approach to running promotions to reduce churn.
This is a sophisticated evolution from the traditional approach to incentivize all customers
equally to reduce churn as it allows companies to spend their marketing budget more
effectively. It is this managerial usefulness of being able to predict churn that attracted us to
this assignment. In this project, we have mainly used classical data prediction techniques of
classification tree and logistic regression to obtain accuracy with error rate of 42%. Taking into
consideration the fact that misclassifying a churner as a non-churner, we lowered the cutoff of
probability of churn for the classification to 0.4. Although this has resulted in a higher rate of
error, it has reduced the overall cost of misclassification, which is the objective of the

Application Area: