A Hybrid Approach CNN-LSTM Based on Attention Mechanism for Credit Card Fraud Detection

Abstract

The increasing use of credit cards has been accompanied by a growing risk of misuse, leading to significant financial losses. Credit card fraud is now recognized as one of the most intricate types of fraud, creating an urgent need for effective prevention methods. This study introduces an innovative fraud detection model that leverages a hybrid approach to address key challenges, such as limited access to public datasets and class imbalance. In this study, we developed a hybrid model combining convolutional neural networks (CNN) and long short-term memory (LSTM) networks enhanced with an attention mechanism to detect fraudulent credit card transactions. The CNN component extracts meaningful features from the data, while LSTM networks identify patterns within fraudulent transaction sequences, and the attention mechanism focuses on critical historical information that might otherwise be overlooked during predictions. Additionally, the synthetic minority oversampling technique (SMOTE) is applied to balance class distribution within the dataset. A comparative analysis was conducted against widely used machine learning and deep learning algorithms, including random forest (RF), support vector machine (SVM), decision tree (DT), multilayer perceptron (MLP), and LSTM-based models. The model underwent extensive empirical evaluation using a European credit card fraud dataset. The results demonstrate that the proposed model excels in detecting fraudulent transactions, achieving an accuracy of 0.9997, precision of 0.9991, recall of 1.000, F1-score of 0.9995, and an area under the curve (AUC) score of 1.000. These findings highlight that the proposed model outperforms current state-of-the-art machine learning and deep learning techniques, establishing it as a robust and effective solution for real-world credit card fraud detection systems.