A Hybrid Graph Neural Network-Based Reinforcement Learning Approach for Adaptive Cybersecurity Risk Management in FinTech

The increasing difficulty of cyber risks in FinTech demands intelligent, adaptive risk management solutions that can learn and develop with dynamic threat environments. Conventional machine learning models frequently fail to adequately capture the structural dependencies and sequential decision-making required to overcome complex attacks. This study introduces an innovative hybrid architecture that integrates Graph Neural Networks (GNNs) with Reinforcement Learning (RL) for dynamic cybersecurity risk management. The GNN component efficiently represents inter-entity interactions in cybersecurity data, whereas the RL module adaptively enhances threat response strategies according to changing system states. A benchmark dataset obtained from Kaggle was utilised to evaluate the system's performance, demonstrating major improvement compared to conventional methodologies. The suggested GNN-RL model attained an accuracy of 96.8 %, precision of 95.2 %, recall of 94.7 %, and a ROC-AUC score of 98.3%, higher than baseline methods such as SVM, Random Forest, and independent GCN. This illustrates the system's capacity to professionally adjust to complicated attack vectors and reduce threats in real time. This research presents a scalable and context-sensitive cybersecurity solution by mixing graph reasoning with reward-based decision-making. The results indicate significant implications for the use of AI-driven cyber defence frameworks in high-stakes financial settings, representing progress in the advancement of autonomous security systems.