Trustworthy and Interpretable Machine Learning Models for Financial Fraud Detection

Increasing computational power, sophisticated algorithms and methods along with advancing technologies have made not only security systems more complex, but also the attack techniques used by malicious actors. Especially in critical areas such as finance, it has become more difficult to detect these sophisticated attacks with high accuracy and at the same time ensure transparency and explainability. In this paper, we develop machine learning (ML) based models for fraud detection on credit card transaction data. By making the developed models explainable using methods such as SHAP, LIME and ELI5, we contribute to both user trust and the sustainability of financial systems. In the experimental results, while achieving 99% accuracy with ML models, we ensured the transparency of the decisions through XAI methods. In this way, we presented an XAI-based approach in the field of financial fraud that offers both high performance and increased user trust.