Improving Predictive Performance in Telecom Churn Modeling with Hybrid SMOTE and GAN-Based Synthetic Data Generation

Customer churn prediction remains one of the most significant challenges in the telecommunications industry, where retaining existing customers is more cost-effective than acquiring new ones. This study proposes a hybrid data augmentation framework integrating Synthetic Minority Oversampling Technique (SMOTE) and Generative Adversarial Networks (GAN) to enhance churn prediction performance by generating high-fidelity synthetic data for the minority class. Six machine learning algorithms—Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), XGBoost, CatBoost, and LightGBM—are implemented and evaluated on both the original and augmented datasets. Traditional machine learning (ML) models often struggle with class imbalance, resulting in reduced sensitivity to churners and lower overall accuracy. The hybrid SMOTE-GAN approach yielded substantial improvements across all performance metrics, with the RF model achieving the highest precision of 85.36%. The results confirm that combining SMOTE and GAN effectively mitigates class imbalance and enhances model robustness, enabling more accurate and actionable customer retention strategies. This framework establishes a scalable and reliable benchmark for data-driven churn management in the telecom industry.