Predicting Customer Churn in the Telecom Sector Utilizing Explainable Artificial Intelligence

Customer churn is a persistent challenge in telecom sector, where retaining present clients is cost-effective than obtaining new ones. Properly churned churn prediction models can deliver valuable decision support to telecom operators. However, conventional machine learning methods are limited by factors like class imbalance, lack of scalability and interpretability, which make them difficult to apply to real-world business environments. To overcome these challenges, this paper presents Explainable Transformer– Ensemble Network (XTENet), a novel customer churn estimate framework in the telecom industry employing explainable artificial intelligence (XAI). The model uses deep learning as the representation learning method and incorporates the XAI methods to enhance clear and business-focused predictions. The proposed framework is compared with the baseline models, such as Random Forest (RF), Logistic Regression (LR), and XGBoost based on publicly available telecom data and generated data representing the network usage patterns. An extended range of metrics is used to assess performance, which includes the measures of classification, imbalance-sensitive, explainability, and business impact. The experimental outcomes show that XTENet has 90% accuracy, 0.81 recall, 0.86 precision, 0.83 F1-score, 0.76 PR-AUC, 0.94 ROC-AUC, 0.85 balanced accuracy, and 0.69 MCC in case of class imbalance. Moreover, the model provides a 0.92 fidelity, 81% stability, 88% sparsity and a 15% retention lift, which is superior to baseline models in both predictive performance and actionable business impact. This study delivers a workable and scalable solution to the task of improving customer retention policies within telecommunication companies by combining predictive power with interpretability.

Перевод пока недоступен