Vol 4 No 2 (2026)

• A Stacking-Based Heterogeneous Ensemble Model for Customer Churn Prediction: Synergistic Integration of LightGBM and AdaBoost

  Ziyu Zeng, Maoxin Li, Yuyi Huang, Zhaohong Cao, Cunjue Li

  Customer churn prediction, typically framed as an imbalanced binary classification problem, poses significant challenges to traditional machine learning models and single ensemble methods, which often suffer from limitations in both predictive accuracy and model interpretability. To address these issues, this paper proposes a heterogeneous ensemble learning framework based on Stacking, which integrates LightGBM and AdaBoost as base learners to leverage their complementary strengths in computational efficiency and classification performance. The proposed model employs a five-fold cross-validation strategy to generate meta-features, thereby enhancing generalization capability. Experimental results demonstrate that the Stacking model achieves an AUC of 0.9132, representing a substantial improvement of 11.45% over standalone LightGBM and 8.69% over AdaBoost. Moreover, the model attains a recall rate of 0.9388, effectively aligning with the business priority of minimizing customer churn through high sensitivity. The innovation of this study lies in three aspects: (1) the design of a heterogeneous ensemble architecture that facilitates performance synergy; (2) the use of cross-validation for robust meta-feature generation; and (3) the incorporation of feature importance analysis to enhance model interpretability. The findings validate the effectiveness of the Stacking ensemble in customer churn prediction and provide both theoretical insights and practical guidance for developing intelligent customer relationship management systems.

  PDF