Implementation of Random Forest Method for Customer Churn Classification

Abstract

Annually, the banking sector consistently undergoes substantial expansion, as demonstrated by the escalating quantity of banks. Nevertheless, this expansion has led to escalating rivalry among banks as they strive to offer superior service to consumers, ultimately impacting customer migration across organizations. Customer churn, or attrition, substantially influences a company's financial performance. Hence, it is crucial to discern the conduct of clients who can discontinue their association with the organization. Precise identification is essential to gather the necessary information for the organization to retain clients and decrease churn rates. An effective strategy for addressing this issue is categorizing client behaviour using historical data. The study utilized the Random Forest approach, employing a 90% training data and 10% testing data ratio. The hyperparameter tuning findings indicate that the optimal parameter combination for constructing a Random Forest model is 400 n_estimators and 40 max_depth. The Synthetic Minority Over-Sampling Technique (SMOTE) mitigates data during categorization. The evaluation of the model demonstrates its exceptional performance in classifying imbalanced data, achieving an accuracy of 90.83%, precision of 89.29%, recall of 92.07%, and  f1-score of 90.66%.