Comparative Evaluation of RNN Architectures for Churn Prediction in Telecom Services

In an era where customer retention is paramount to business sustainability, accurate customer churn prediction is vital for industries like telecommunications. This research investigates the efficacy of three neural network architectures—Simple Neural Network (SNN), Gated Recurrent Unit (GRU), and Variational Recurrent Neural Network (VRNN)—for predicting customer churn. We employ a pre-processed telecom dataset with extensive feature engineering and apply state-of-the-art techniques such as SMOTE to address class imbalance. The models are evaluated based on key performance metrics, including accuracy, precision, recall, and AUC-ROC. Results indicate that while the GRU outperforms the SNN regarding recall and f1-score, the VRNN offers superior latent representation learning, delivering higher accuracy and AUC. These findings underscore the importance of model selection when optimizing churn prediction for enhanced customer retention strategies.