Attention-Based Deep Learning Models for Fraud Detection in Imbalanced Transaction Datasets

The rise of credit card payments in the online industry has led to a significant increase in fraudulent activities, causing substantial financial losses for businesses and institutions. This situation underscores the pressing need for modern artificial intelligence (AI) systems that can automate fraud detection and accurately distinguish between legitimate and fraudulent transactions. Fraud detection is often treated as a binary classification task, focusing on accurately identifying suspicious activities. By incorporating deep learning models with attention mechanisms, this process can be greatly improved. Attention mechanisms help the model focus on the most important features in transaction data, boosting the accuracy and efficiency of detecting fraudulent behavior. In this study, firstly two different data augmentation techniques such as ADASYN and CTGAN are used for solving class imbalance problems in the training sets. Then, we enhance the field of fraud detection by leveraging and contrasting four sophisticated models. The results confirm the strength of these attention-based models. These groundbreaking advancements not only inspire greater confidence in AI systems but also significantly improve the ability to detect fraud in transactions.