Credit card fraud detection has become a critical challenge for financial institutions due to the increasing prevalence of fraudulent activities in digital transactions. This study proposes a novel hybrid model that integrates ResNet for spatial feature extraction, Long Short-Term Memory (LSTM) networks for capturing temporal dependencies, and an Attention Mechanism to prioritize significant features. The model addresses key challenges such as class imbalance, scalability, and adaptability to evolving fraud patterns. Using the IEEE-CIS fraud detection dataset, the study demonstrates significant improvements in fraud detection performance. Synthetic Minority Oversampling (SMOTE) is applied to balance the dataset, ensuring the model effectively identifies rare fraudulent transactions while reducing false positives and negatives. Comparative analysis shows that the proposed framework achieves superior results, including a precision of 96%, recall of 92%, and an F1-score of 93.97%, outperforming benchmark models by a significant margin. The integration of attention mechanisms enhances interpretability, while advanced evaluation metrics like Shapley Additive explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) provide insights into the model's decision-making process. The findings highlight the proposed model's potential as a robust, scalable, and interpretable solution for real-world credit card fraud detection. Recommendations for future research include expanding validation across diverse datasets, exploring advanced architecture like Transformers, and enhancing computational efficiency for real-time deployment. This study establishes a strong foundation for improving fraud detection systems and contributes to advancing machine learning methodologies in financial security applications.
Umaru et al. (Wed,) studied this question.