Hybrid CNN-LSTM With Attention Mechanism for Robust Credit Card Fraud Detection

Abstract

In an era marked by rapid technological advancements and a global shift toward cashless transactions, credit card fraud has emerged as a significant challenge, causing substantial financial losses and threatening the security of consumers and financial institutions. The exponential growth of online payments necessitates more effective fraud detection mechanisms. Traditional fraud detection methods, including rule-based systems and basic statistical analyses, struggle to keep pace with the evolving strategies employed by fraudsters. This paper proposes a hybrid fraud detection model integrating Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and an attention mechanism to address these challenges. Each component of the hybrid model is designed to capture specific behavioral representations, with CNNs focusing on spatial features, LSTMs handling temporal sequences, and the attention mechanism highlighting the most relevant features. We utilized a benchmark dataset and applied the Synthetic Minority Over-sampling Technique (SMOTE) to balance the class distribution. Extensive data preprocessing was conducted to ensure compatibility with the model’s input requirements. The experimental results demonstrated that our hybrid model significantly outperforms traditional machine learning algorithms, achieving an accuracy of 99.93% and a recall of 0.89 on the European cardholders’ dataset. These results highlight the model’s ability to detect fraudulent activities with high precision and reliability. Our hybrid CNN-LSTM-Attention model improves fraud detection by addressing both spatial and temporal data features while dynamically focusing on critical elements. This work not only contributes to the advancement of fraud detection techniques but also provides a framework for future research.

Topics & Concepts