Telco Customer Churn Prediction Using Composite Deep Learning Technique

Customer churn is a critical issue in the telecommunication (Telecom) industry, leading to significant revenue loss and increased customer acquisition costs. To address this, a project was initiated to reduce home broadband (HBB) postpaid churn from 20% within three months. This study introduces novel hybrid deep learning (DL) models, such as BiLSTM, CNN, and BiLSTM+CNN, which combine bidirectionallonglshortterm memory (BiLSTM) with Convolutional Neural Network (CNN) to enhance churn prediction accuracy. Additionally, the study explores the significance of feature selection in churn prediction and the potential of DL to automate this process, eliminating the need for expert intervention. Experimental results reveal that the BiLSTM model achieves a notable 76% precision when trained, tested, and validated on benchmark datasets, outperforming other DL techniques in terms of accuracy, precision, recall, and F1-score. The study emphasizes the importance of data pre-processing and feature engineering, highlighting the significance of analyzing data usage patterns and household connections. Moreover, it suggests actionable strategies for Telecom companies to mitigate churn, such as targeted campaigns based on predicted churn reasons.