The head of the SME division had asked whether it’s possible to predict the customers which were most likely to churn so that they could trial a range of preemptive actions.
Customer churn prediction is a very serious problem for the company, so we can’t proceed with naive ways like comparing different classifiers based on accuracy. First we cleaned the data and applied feature engineering to make data appropriate for the task and classier. Then we created a cost matrix based on our field knowledge and discussion with the client. This cost matrix was then linked with the machine learning classifiers to identify their classification accuracy.
With the help of Machine Learning algorithms, we can reduce the costs from $107 to $97 per customer because there was 16000+ customers in this company.