An Enhanced Framework for Churn Prediction using Stratifed Bagging and Stacked Multi-Output Random Forests

Customer churn prediction is an important task in customer relationship management because it helps businesses know who is at risk of leaving and retain such at-risk customers.Accurate and time-efficient churn prediction is essential to identify the factors driving churn, enabling businesses to implement timely and effective retention strategies.A framework is proposed which combines stratified bagging with multi-output Random Forest and stacked feature transformation, to enhance prediction accuracy. This approach balances the dataset, trains a custom Random Forest using stratified bagging, and applies stacked learning using Logistic Regression on transformed features. Tested on real-world data, the model demonstrated 99 % accuracy, showing improvements over the classic Random Forest in precision, recall, and prediction time.