Interpretable Credit Card Fraud Detection Using LightGBM and XAI Techniques

Credit card fraud presents an important challenge for the financial industry, made very difficult from the constantly evolving nature of fraudsters strategies. The challenges concerning fraud detection are multiplied by the transaction data which is very imbalanced with fraudulent transactions accounting for only 0.17% of the data. The evidence-based developments of machine learning have helped the detection of fraud but there is a growing need for transparency with reliable interpretable decision processes in financial systems. This study investigates a fraud detection system based on the LightGBM classifier alongside explainable AI (XAI) methods using the Kaggle Credit Card Fraud dataset containing 284,807 transactions and identifying 98.4% accuracy, against competing approaches. The XAI methods particularly LIME, provides clarity of the features driving the prediction and explains why the transaction was flagged. The practical use of XAI and fraud detection algorithms is that they combine complicated models in a format that has defended transparent justification of outcomes. Therefore, this study effectively combines accuracy and transparency. This added dimension helps enhance fraud detection systems and supports a global change in how risk is evaluated and customer protection is delivered in a financial context. Overall outcomes provide evidence and rationale for (AI) systems to be trusted in fraud prevention systems.