AI-BASED CUSTOMER CHURN PREDICTION WITH EXPLAINABILITY

This study examined the use of machine learning and Explainable Artificial Intelligence (XAI) for customer churn prediction, with a focus on both predictive performance and the interpretability of model decisions. Five machine learning models—Logistic Regression, Decision Tree, Random Forest, XGBoost, and LightGBM—were evaluated using multiple performance measures, including Accuracy, Precision, Recall, F1-Score, ROC-AUC, and PR-AUC. SHAP and LIME were additionally used to understand the factors contributing to churn predictions at global and individual levels.   Among the evaluated models, Logistic Regression provided the strongest overall performance according to the criteria considered in this study. It achieved a ROC-AUC of 0.8520 and a Recall of 0.7219, while its Accuracy of 0.8028 was equal to that of LightGBM. LightGBM produced the highest Precision, XGBoost achieved the highest PR-AUC, and Decision Tree recorded the highest F1-Score. These results demonstrate that model performance can vary depending on the evaluation measure, making it important to consider multiple metrics rather than relying on a single measure.   The explainability analysis provided additional insight into the factors associated with churn predictions. SHAP identified customer-related characteristics such as tenure, monthly charges, total charges, contract type, internet service, and other service-related attributes as important contributors to model predictions. LIME further demonstrated how these factors influenced an individual high-risk customer prediction. Although several important factors appeared across different models and explanation techniques, their relative contribution was not identical, indicating partial rather than complete consistency.   The findings therefore suggest that selecting a churn prediction model should not depend only on predictive performance. Interpretability is also valuable when model predictions are intended to support customer retention decisions. A model may provide strong predictive results while another model or explanation technique may offer different insights into the reasons for churn. Combining predictive evaluation with SHAP and LIME can provide a broader view of both what the model predicts and why it makes a particular prediction.   Overall, this study demonstrates that machine learning-based churn prediction can be complemented by XAI to produce more transparent and informative decision support. The combined use of predictive performance measures and explanation techniques can help organizations better understand customer churn patterns and make more informed retention decisions. Further research using additional datasets, models, and XAI techniques can help assess the generalizability and stability of these findings across different customer and business environments.

[1]    K. Peng, Y. Peng, and W. Li, “Research on customer churn prediction and model interpretability analysis,” PLOS ONE, vol. 18, no. 12, p. e0289724, Dec. 2023, doi: 10.1371/journal.pone.0289724.  [2]   U. Tiwari, S. Ashwani, A. J. Tripathy, and K. D. Kumar, “A two-stage ensemble approach for analysis of optimizing customer churn with LIME interpretability,” in 2024 IEEE International  Conference  on  Computing,  Power  and Communication Technologies (IC2PCT), 2024, pp. 1540– 1545, doi: 10.1109/IC2PCT60090.2024.10486713.  [3]   “Explaining customer churn prediction in telecom industry using tabular machine learning models,” Machine Learning with Applications, vol. 17, p. 100567, Sep. 2024, doi: 10.1016/j.mlwa.2024.100567. [4]      “Hybrid black-box classification for customer churn prediction with segmented interpretability analysis,” Decision Support Systems, vol. 181, p. 114217, Jun. 2024, doi: 10.1016/j.dss.2024.114217.  [5]   A. Manzoor, M. A. Qureshi, E. Kidney, and L. Longo, “A review on machine learning methods for customer churn prediction and recommendations for business practitioners,” Journal of Business Research, 2024.  [6]        “Prediction of customer churn behavior in the telecommunication industry using machine learning models,” Algorithms, vol. 17, no. 6, p. 231, 2024.  [7]     “A comprehensive evaluation of machine learning and deep learning models for churn prediction,” Information, vol. 16, no. 7, p. 537, 2025.  [8]      A. El Attar and R. El-Hajj, “Explainable AI-driven customer churn prediction: A multi-model ensemble approach with SHAP-based feature analysis,” Frontiers in Artificial Intelligence, vol. 9, 2026, doi: 10.3389/frai.2026.1748799.  [9]    S. M. Lundberg and S.-I. Lee, “A unified approach to interpreting model predictions,” in Advances in Neural Information Processing Systems (NeurIPS), vol. 30, 2017.  [10]  M. T. Ribeiro, S. Singh, and C. Guestrin, “‘Why should I trust you?’: Explaining the predictions of any classifier,” in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016, pp. 1135–1144.