CNN-BiLSTM hybrid model for customer churn prediction in Cell2Cell

Predicting customer turnover is a crucial problem in the telecom sector, because keeping existing customers is often less expensive than finding new ones. Suboptimal prediction performance results from standard machine learning algorithms' inability to identify both spatial and sequential patterns in consumer behaviour. In order to improve churn prediction, we propose an innovative hybrid model in this study that combines CNN and BiLSTM networks. While the BiLSTM layer records temporal dependencies and sequential patterns in customer behaviour, the CNN component extracts high-level spatial characteristics from consumer contact data. A telecom dataset that includes occurrences of premature, involuntary, and consensual churn is used to train and assess the model. The CNN-BiLSTM hybrid model performs better than both solo deep learning and traditional machine learning methods, according to experimental result.