Bagging and Boosting for Predicting Bank Customer Churn

Customer churn in a bank impacts its profit, costs, and reputation. Retaining consumers is also less expensive and more profitable than acquiring new ones. In order to retain a customer, it is vital to establish an approach to predicting whether or not the customer will churn. Machine learning models like bagging and boosting will help with this. Bagging and boosting are appropriate for dealing with this since these approaches can deal with imbalanced data well, and data imbalances are common in churn cases. The dataset in this study is also imbalanced, with 7,963 out of 10,000 customers not churning. The two bagging models, Random Forest and Extra Trees, and the two boosting models, Gradient Boosting and XGBoost, are built after the dataset is processed and balanced using SMOTE. These four models’ performances will be compared both before and after tuning. The results showed that tuned XGBoost was the best model in this study, with an AUC, accuracy, precision, recall, and f1 score of 0.786, 0.861, 0.642, 0.662, and 0.652, respectively. In this study, boosting is more effective than bagging. The "NumOfProducts", "Geography", and "Age" attributes have the most impact on the best model’s prediction. Therefore, banks can use the results of this study as a decision-making tool.