Abstract:

Customer churn prediction is a well-known application of machine learning and data mining in Customer Relationship Management, which allows a company to predict the probability of its customer churning. In this study, we extended the application of customer churn prediction to the context of software maintenance contract. In addition, we examined the predictive power of economic factors. Random forest, gradient boosting machine, stacking of random forest and gradient boosting machine, XGBoost, and long short term memory networks were applied. While an ensemble model and XGBoost performed best, macroeconomic variables did not yield statistically significant improvement in any prediction.