Credit Card Fraud Detection using TabNet

The adopting of cashless payment methods, such as credit card payments and online transactions, has significantly enhanced the payment experience and added convenience to our daily lives. However, with the increase in cashless payment usage, financial fraud has become more sophisticated, posing a significant challenge to the security of these payment systems. In response, machine learning-based approaches have gained popularity in fraud detection. In this research paper, we propose the application of a deep tabular learning model, TabNet, for classifying transactions into fraudulent or non-fraudulent categories. TabNet utilizes a sequential attention mechanism to learn from tabular data. It comprises a series of decision steps where each step selects relevant features and updates the internal representation of the data. This mechanism enables the model to effectively capture complex, non-linear relationships between features, making it highly effective for fraud detection. The utilization of TabNet in fraud detection can contribute to strengthening the security of the payment system and reduce the chance of financial fraud. To evaluate the efficacy of our proposed approach, we conducted experiments on three widely used credit card fraud datasets, including the MLG-ULB dataset, the IEEE-CIS Fraud dataset, and the 10M dataset. Our experiments revealed that TabNet outperforms the state-of-the-art approaches across all three datasets, demonstrating its robustness and effectiveness in detecting fraudulent transactions.