Application Design of Customer Churn Prediction Using Random Forest and XGBoost Algorithms for Telecommunication Industry in Indonesia

Customer churn is a primary challenge in the telecommunications industry, directly impacting revenue and increasing acquisition costs. This research designs a churn prediction system for Indonesian ISP services using Machine Learning algorithms, specifically Random Forest and XGBoost. To address the prevalent issue of class imbalance in churn data, we implement a hybrid approach combining Cluster-Based Undersampling with Cost-Sensitive Learning. Explainable AI (XAI) methods were applied to interpret model predictions, specifically LIME and SHAP, to provide transparent interpretations of model predictions at both global and local levels. Customer segmentation using K-Means clustering is integrated to support personalized retention strategies. The final output is an interactive dashboard built with Streamlit, serving as a decision-support tool for management. Our results show that the XGBoost model outperforms others with a ROC-AUC of 0.900 and Recall of 0.907. The study includes a discussion on the business impact and limitations, highlighting the potential for significant cost savings through proactive retention.