Explainable AI Framework Brings Churn Prediction into Telecom CRM

Summary

A new arXiv paper proposes a four-layer CRM architecture that integrates telecom churn classifiers with SHAP and LIME explanations. Logistic Regression achieved the best AUC-ROC, while LightGBM had the highest accuracy. The system turns risk scores and feature attributions into customer segments and retention templates, with campaign results returned to model retraining.