Advanced Customer Churn Prediction using CNN-LSTM Hybrid Model and Ensemble Voting Mechanism

Abstract

In the telecom sector, where keeping current clients is more economical than finding new ones, customer churn prediction is crucial. In order to efficiently capture both spatial feature patterns and temporal relationships in consumer behavior data, this research suggests an advanced hybrid architecture that integrates Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM) networks. To improve prediction accuracy and resilience, an ensemble Voting Classifier that integrates Boosted Decision Trees and Extra Trees is also used. SMOTE-based data balancing strategies are used to solve the problem of class imbalance, guaranteeing equitable and dependable model performance. Additionally, a Flask-based web application is created to offer a user-friendly interface for safe access and real-time churn prediction. In terms of accuracy, generalization, and stability, experimental assessment shows that the suggested hybrid technique performs noticeably better than standalone deep learning models and conventional machine learning. Index terms - — Customer Churn Prediction, CNNLSTM, Ensemble Learning, Voting Classifier, SMOTE, Class Imbalance, Deep Learning, Telecommunication Industry, Flask