Customer Churn Prediction in the Telecommunication Industry to Support Customer Retention Strategy Using the XGBoost Algorithm

Abstract

The telecommunications industry faces significant challenges regarding customer churn, as the cost of acquiring new customers is substantially higher than retaining existing ones. This research aims to build an optimal customer churn prediction model using the XGBoost algorithm to support customer retention strategies. The methodology employed is the Cross-Industry Standard Process for Data Mining (CRISP-DM), utilizing a secondary dataset from Kaggle comprising 7,043 customer records. The data imbalance challenge was addressed using the Synthetic Minority Over-sampling Technique (SMOTE) on training data to enhance minority class detection. Evaluation results using Stratified 5-Fold Cross Validation demonstrate highly reliable model performance, achieving an accuracy and a recall value. This model was subsequently implemented into a web-based information system using the FastAPI framework and Neon Database. The system is capable of performing real-time customer risk classification (Low, Medium, High) and integrates Generative AI technology (Gemini API) to provide personalized and communicative retention strategy recommendations for customer service agents.