PREDICTION OF CUSTOMER CHURN USING MACHINE LEARNING

Abstract

Companies place great emphasis on Customer Relations Management, in particular the factors related to customer retention, and, conversely, the rate of "churn", the loss and replacement of customers. The prediction of churn is especially important, due to the economic advantages of retaining and satisfying existing customers over the costs of acquiring new ones. Despite a plethora of research dedicated to the topic, it remains a challenge for commercial enterprises to accurately predict customer churn.

Machine learning, and more recently deep learning, have emerged as effective tools for the analysis of client data to help identify relevant factors and predict rates of retention and churn. Commonly employed methods include Random Forest, Gradient Boosting, ANN, XGBoost, Decision Trees, Support Vector Machine, Adaptive DNN, and MLP hybrid classifiers.

In this study, we analyze multiple open-source customer datasets using machine learning methods, based on the literature. We carefully select datasets with varying characteristics and from different domains, and apply best-performing algorithms to predict customer churn. We have successfully replicated previously published work along with some variations described in the text.