An Explainable TabNet-Based Approach for Credit Card Fraud Detection

The application of noncash payment methods, including credit card payments and online transactions, has contributed to improving user experience and bringing convenience to daily life. However, along with the popularity of these methods, financial fraud is becoming increasingly sophisticated, posing a major challenge to the security of the payment system. In this context, machine learning-based methods are being increasingly applied in fraud detection. In this study, we propose applying the TabNet deep learning model to classify tabular transactions as either valid or fraudulent. TabNet uses a sequential attention mechanism to learn from tabular data through a series of decision-making steps. At each step, the model selects important features and updates the internal representation of the data. Thanks to this mechanism, TabNet can effectively exploit complex and nonlinear relationships between features, thereby improving the accuracy in detecting fraudulent transactions.