Graph-Temporal Hybrid Neural Networks for Enhanced Malware Detection in Banking Systems

As the banking sector faces an increasing threat from sophisticated malware attacks, traditional detection methods struggle to keep pace, resulting in high false positive rates and missed detections of emerging threats. This paper presents a novel Graph-Temporal Hybrid Neural Network (GTHNN) framework that combines advanced graph neural networks with temporal analysis to improve the accuracy and resilience of malware detection systems in financial institutions. The GTHNN model is designed to effectively analyze complex relationships and evolving patterns within both malware behaviors and user interactions by leveraging graph structures and sequential data.A key innovation of this research is the development of a domain-specific dataset that includes a diverse range of real-world malware samples and legitimate banking transactions, enabling the model to adapt to previously unseen threats, including zero-day attacks. Extensive performance evaluations conducted in simulated banking environments demonstrate that the GTHNN framework significantly outperforms existing detection models with 96.8% accuracy with low false rate. Furthermore, the duallayered approach effectively minimizes operational disruptions by reducing false positives.