Explainable AI-Based Fraud Transaction Detection using XGBoost and SHAP

Abstract: The quick development of online payment systems and digital banking has contributed to a significant increase in financial fraud,resulting in significant economic losses and present serious threats to the security of digital finanacial services.Machine learning is being developed to enhance fraud-detection accuracy,yet many current methods remain limited to opaque,black-box models which may reduce user trust in automated decisions.In this paper a framework is presented that identifies fraud transaction patterns through the application of Explainable Artificial Intelligence (XAI) to provide correct classification as well as interpretable predictions.It involves a risk-oriented feature engineering approach that detects hidden fraud patterns,and a complete learning model capable of flagging suspicious transactions while offering detailed insight into the factors driving every prediction.Experiments evidence indicate the proposed method has the capability to reliably and safely identify fraudulent transactions, achieving an overall accuracy of 97.61%.Moreover,the explainability feature supports the transparency by highlighting the influence of key features to each prediction,which makes the framework suitable for reliable and practical applications of financial fraud detection.

Keywords: Fraud Transaction Detection; Explainable Artificial Intelligence (XAI); XGBoost; SHAP; Machine Learning; Feature Engineering; Banking Transactions; Financial Security.