An Improved Churn Prediction Model using Relevance Feature Identification and Personalized Modeling Technique with Optimized Retention Strategies

Churn prediction is vital in customer retention strategies, helping businesses proactively address customer dissatisfaction. In this research, we propose a comprehensive churn prediction model that addresses these challenges by integrating advanced techniques, including handling imbalanced data, feature relevance identification, new feature generation, incorporating temporal patterns, and personalized modeling for customer segments. By applying advanced techniques such as SMOTE-ENN for balancing, RFE-RF for feature elimination, and K-Means clustering for segmentation, the model improves predictive accuracy and provides actionable insights for targeted retention strategies. The proposed model effectively enhances prediction accuracy and provides actionable insights for customer groups, optimizing retention strategies. With the proposed Model Logistic Regression performs better across all segments in terms of AUC-ROC and precision particularly for high-value and frequent users with $0.91 \%$ and $0.80 \%$ respectively.