Customer Churn Prediction and Recommendation using an Intelligent System

Abstract

Customer churn remains a critical challenge in the retail sector, often resulting in significant revenue loss and increased acquisition costs. This paper presents an integrated deep learning framework that combines unsupervised clustering, predictive modeling, and rule-based recommendation to identify and retain at-risk customers. The approach begins with customer segmentation using KMeans clustering enhanced by Principal Component Analysis (PCA) for reduction of dimensional complexity of the dataset, followed by segment-specific churn prediction using a hybrid BiLSTM–CNN architecture. This model leverages temporal patterns and local feature extraction to improve prediction accuracy. Based on the predicted churn probability, a rule- based recommendation engine issues targeted digital coupons or retention strategies, enhancing business decision-making. The entire workflow is deployed through a user-friendly Streamlit web application, enabling real-time data analysis, visualization, and strategic action. Experimental results demonstrate that segment-specific models outperform global models, and the rule-based system offers interpretable, customizable recommendations. The proposed system is scalable, domain-adaptable, and effective in reducing churn while supporting informed, cost-efficient marketing interventions.