Klasifikasi Customer Churn Menggunakan XGBoost dengan Optimasi GridSearchCV Berbasis Shapley Additive Explanations

Abstract

Customer churn is a significant challenge in the banking sector, often leading to revenue loss and requiring predictive strategies to enhance customer retention. This study implements the Extreme Gradient Boosting (XGBoost) algorithm for churn classification, with hyperparameter optimization using the GridSearchCV technique to improve model performance. The dataset comprises 10,000 banking customers with 9 features and 1 target label. To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied. Prior to tuning, the XGBoost model achieved an accuracy of 80.8%. After applying optimal parameters, the model's performance improved to 81.5%, along with higher precision and recall values, indicating improved robustness and consistency. For model interpretability, Shapley Additive Explanations (SHAP) were used and visualized through a beeswarm Plot. The analysis identified age, customer activity status, and number of products owned as the most influential features in predicting churn. Based on these findings, this study proposes business recommendations including age-based customer segmentation, enhancing active customer engagement, and optimizing product offerings as strategies to reduce churn.