Customer Churn Prediction on Structured Data Using FT-Transformer and Stacking Ensembles

In three linesarXiv paper introducing hybrid FT-Transformer + XGBoost architecture with calibration-aware stacking for customer churn prediction on structured data. Achieves 62.10% F1 and 0.861 AUC-ROC on public bank dataset, outperforming MLP baseline by 3.37 F1 points. Handles class imbalance using class-weighted loss without synthetic oversampling.

Summary generated by Claude — human-verified