This project, I have done as a part of an interview process for a popular e-commerce site. The main goal was to predict the customer churn based on given few features.
Note: The original data is not revealed due to privacy concerns.
In summary I have buit 6 models to compare the predictive power in each one of them. I have selected 3 models created using logictic regression, SVC and Neural Network. The main selection criteria was the lowest False Negative rate of the results, since the higher the value of this means the model is not capable of correctly identifying targets. At the same time False Positive rate needed to be as low as possible. Because there will be plans to launch marketing programs to retain the customers who are predicted as highly probable to churn with these models. And the company might wanted to offer discounts for these target customers and they do not want to incur any cost on the customers who are not likely to churn.