Asian Journal of Research in Computer Science

Customer churn remains a critical challenge for the telecommunication industry, directly impacting revenue and customer retention strategies. Traditional churn prediction models based on statistical and machine learning techniques have shown limited adaptability in capturing complex behavioural patterns. Deep learning (DL) methods, particularly recurrent neural networks (RNN), convolutional neural networks (CNN), and transformer-based architectures, have emerged as powerful tools for modelling customer churn by leveraging vast and dynamic datasets. By enabling precise identification of at-risk customers and personalized intervention strategies, DL-driven approaches not only enhance retention rates but also unlock targeted revenue-generating opportunities through tailored service upgrades and dynamic pricing models. This paper presents a comprehensive review of deep learning-based churn prediction mechanisms in the telecommunication sector, comparing their architectures, feature engineering strategies, and performance metrics. Highlights of recent advances in DL techniques, including attention mechanisms and explainable AI, were presented and their implications for improving customer retention strategies were discussed. Finally, key research challenges and future directions—such as developing DL models with simpler explainability frameworks, advancing techniques for class imbalance mitigation, and designing adaptive architectures for real-time, resource-efficient inference—were highlighted to bridge the gap between theoretical innovation and scalable deployment in operational environments.