Application of Machine Learning Algorithms to Predict Customer Churn in E-commerce Platforms

Abstract

The amount of e-commerce data available is growing exponentially and small business owners are desperate to make sense of their user, product, and transaction data. With increasing levels of competition, businesses are particularly interested in knowing if and when they will lose customers, otherwise known as customer churn. In this thesis, we create a customer purchase history data set using data obtained from a small ecommerce mobile application based in Buenos Aires, Argentina. We aim to predict if a customer will leave for a competitor (churn) or not (non-churn) based on their purchase history. We apply support vector machine and logistic regression models to our data set and find that both models successfully distinguish between the two classes of customers.