Predicting Bank Customer Churn
Using machine learning to flag at-risk customers and inform retention strategy.
Result: Random Forest was the best model, with 95.39% accuracy and an AUC of 0.98. Transaction amount and transaction count were the strongest churn signals.
Goal
Predict which bank customers are likely to churn so the bank can retain them before they leave. The analysis also feeds into broader customer relationship management decisions.
Approach
- EDA: Explored customer demographics, transaction behavior and financial habits.
- Preprocessing: Applied feature scaling and categorical encoding.
- Modeling: Trained Random Forest, SVM, Decision Tree, kNN and LASSO logistic regression.
- Evaluation: Compared models on accuracy, precision, recall and AUC.
Takeaways
Across every model, Total_Trans_Amt and Total_Trans_Ct were the strongest churn signals. Customers who transact less are the ones to target with retention campaigns, which makes a clear case for data-driven outreach.