Optimizing Credit Card Fraud Detection Using Machine Learning and Explainable AI

Abstract

Following the booming development of e-commerce and online payment systems, credit card fraud has become a major issue of concern among financial institutions and consumers. But that is exacerbated by the fact that fraudulent transactions constitute less than 0.2 percent of all transactions. This paper showcases the use of Machine Learning (ML) models to detect credit card frauds. The paper uses four ML algorithms and the Synthetic Minority Oversampling Technique (SMOTE) to balance out the data for fraud cases. ML models like Logistic Regression, XGBoost, Random Forest and LightGBM were tested on the data. Models were made better by cleaning up the data-scaling features and resampling of the data. The results show that XGBoost is more efficient at finding patterns in fraud cases with an accuracy of 97 percent, precision of 97 percent and recall of 93 percent. The study says that using models like XGBoost with oversampling techniques can really improve fraud detection. Also, the model can explain its predictions using Explainable AI algorithms like SHAP and LIME. This enables to see how the model is finding patterns. SMOTE is employed to balance the data and Explainable AI for understanding the algorithm, resulting in model efficiency. These models are highly accurate and is suitable in real-world situations to detect financial fraud. Overall, the proposed system is effective and usable for real-life finance fraud prevention along with data balancing (SMOTE) and interpretability (XAI).