Churn Prediction Models Using Gradient Boosted Tree and Random Forest Classifiers

Abstract

In the era of a competitive market, every organization has been used a lot of marketing techniques to maximize their profit and to preserve the existing flow of customer relationships with the firm. The cost of attracting a new customer incurs more times than retaining existing ones. Thus, customer relationship management (CRM) analyzers try to know the behavior of customers and find the causes of a customer churning. To produce a list of telecom customers who likely to churn in the future, this paper presents the two churn prediction models using wrapper-based Forward Feature Selection (FFS) with Gradient Boosted Tree and Random Forest classifiers. This work analyzes the FFS with five comparative classifier models based on the telecom data using the KNIME analytics platform and deploys the two most accurate models with a new dataset to predict the future churn. Our models achieve the accuracies of 96.2% and 96.89% respectively.