Machine Learning Models for Bank Customer Churn Prediction: A Comparative Study of LightGBM, CatBoost, and XGBoost

This research is dedicated to forecasting bank customer churn by using machine learning methods. In the highly competitive landscape of the banking industry, retaining customers has emerged as a critical factor for a bank's success. Eleven distinct machine learning models were utilized in the analysis of a bank customer dataset. Prior to model implementation, comprehensive data preprocessing was conducted, encompassing the management of missing values, encoding of categorical variables, and detection of outliers. The models were then evaluated under various data split ratios and sampling methods, with recall serving as the primary evaluation metric.The study's findings indicate that LightGBM, CatBoost, and XGBoost demonstrate superior performance in predicting customer churn. Additionally, Gradient Boost, Ada Boost, and Bagging prove to be effective in handling imbalanced data. Several key predictors of customer churn, such as Total_Trans_Amt and Total Trans ct, were identified. These results offer valuable practical implications for banks in formulating customer retention strategies. However, it is necessary to recognize the research's limitations. There may be deficiencies in data representation. What's more, external factors that could influence customer churn have been excluded.