This study introduces a novel framework for detecting fraud in digital banking, particularly addressing challenges like imbalanced data and operational limits. The framework incorporates chronological evaluation, feature engineering to prevent data leakage, probability calibration, and SHAP for explainability. Experiments on the PaySim benchmark showed that XGBoost performed best among tested models, achieving a PR-AUC of 0.5658. The research also highlighted a trade-off between reducing false positives and maintaining recall when implementing an adaptive policy. AI
IMPACT Provides a methodological framework for improving fraud detection in financial services, potentially leading to more secure digital banking operations.
RANK_REASON Academic paper detailing a new machine learning framework and experimental results. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
