Explainable Customer Churn Prediction Using Pseudo Temporal LSTM Modeling, SHAP Based Attribution and Conversational AI Decision Support

This paper presents an intelligent customer churn prediction framework designed for subscription based businesses. The proposed system combines a pseudo temporal Long Short Term Memory (LSTM) model with SHAP based explainability and a conversational decision support interface to improve both prediction accuracy and interpretability. Since the IBM Telco Customer Churn dataset does not contain sequential information, synthetic temporal sequences are generated to enable sequence based learning. In addition to predicting whether a customer is likely to churn, the framework also identifies the major factors influencing churn and presents the insights through a user friendly conversational interface, making the system more practical for real world business decision making.