A Comprehensive Evaluation of Machine Learning and Deep Learning Models for Churn Prediction

Nabil M. AbdelAziz, Mostafa Bekheet, Ahmad Salah, Nissreen El-Saber, Wafaa T. AbdelMoneim

Churn prediction has become one of the core concepts in customer relationship management within the insurances, telecom, and internet service provider industries, which is essential in customer retention. Therefore, this study attempts to analyze the effectiveness of the advanced machine learning and deep learning models for churn prediction in the evaluation of the models’ performance across different sectors. This would help conclude whether the varied patterns of the churn throughout different sectors to the level that affects the model performance and to what extent. The work includes three datasets: namely, insurance churn, internet service provider customer churn, and Telecom churn datasets. The implementation and comparison conducted in this study of models include XGBoost, Convolutional Neural Networks (CNNs), and Ensemble Deep Learning with the pre-trained hybrid approach. The results show that the ensemble deep learning model outperforms other models in terms of accuracy and F1-score, achieving accuracies of up to 95.96% in the insurance churn dataset and of 98.42% in the telecom churn dataset. Moreover, traditional machine learning models like XGBoost also produced competitive results for selected datasets. The proposed deep learning ensembles reveal the strength and possibility for churn prediction and provide a benchmark for future research relevant to customer retention strategies. Also, the proposed ensemble deep learning model shows stable performance across different sectors, which reflects its ability to capture the varied churn patterns of different sectors.