Churn Prediction using Deep Neural Network

Customer churn is essential in business as getting new customers requires more efforts and investment. This business logic is also applicable for retaining other business stake holders like employees, suppliers, investors etc. Whether it is getting new clients, hiring new employees, sourcing new suppliers or getting new investors, they all involve additional costs and investments. This article is about training a deep neural network to predict purchase behavior of customers for an application selling digital books.

Here is the brief overview of the pipeline -

The image on the right describes the architecture of the model. The model has two hidden layers, with relu as activation function, adam used as optimizer, trained with batch size 100 over 100 epochs.

Input variables here include average number of books purchased, sum of books purchased, average price paid, sum of price paid, review (boolean, given or not), review(int value max 10), total pages read, completion percent (value between 0 and 1), number of support requests, number of days last visit since day of purchase. The prediction is of how the customer will convert within a span of 6 months, based on customer engagement activity of past 2 years.

The model can be used in two ways, a) to understand the probability of a customer converting, b) predict whether they buy (1) or not (0).

Please share your feedback and suggestions in the comments section.

Happy learning!