Problem Use Case
One of the leading Telecom companies was facing massive losses and so wanted to predict the customer churn ahead of time so that they could come up with some attractive offers for the people who are likely to churn.
The customer provided us with data which described the number of customers who churned the service previosly.
It consists of 3333 observations having 21 variables.
We created a Random Forest model, having 2850 observations as level 0(Non-Churner) and 483 observations as
level 1(Churner) which shows class imbalance.
In the confusion matrix, for class 0(non-churner) we had error percentage of 13% and 14% for class 1(churner).
From the variable importance plot, we got to know that Day.Mins, Custserv.Calls, Eve.Mins, Intl.Mins, Int.l.Plan
explain more variance in the model.
- We found that customer who spent more than 200 minutes on day call are more likely to churn.
- The customers who made than more than 3 customer service call were churned more so they are more likely to churn.
- The customer who opt international plan is more likely to churn.
The customer service call , minutes spent on day calls, international calls, evening calls, & night calls and international plan opt by customer were important attributes for predicting Churn.