A Reinforcement Learning Approach to Cybersecurity: Deep Q-Networks for Threat Modeling
Harish Janardhanan · 2025
As the frequency and complexity of cyber-attacks continue to rise, cybersecurity methods struggle to keep pace with evolving threats. This study presents an advanced approach to cybersecurity threat modelling using reinforcement learning. The proposed method integrates Deep Q-Network (DQN) to dynamically allocate cybersecurity resources, enabling continuous adaptation to emerging threats. We utilize a comprehensive dataset consisting of real-world threat actions and cybersecurity capabilities, facilitating the modelling of adversarial behaviors and the evaluation of defense mechanisms. By applying this approach, the DQN model learns to allocate resources efficiently, improving the overall defense of the system in a dynamic environment. The results demonstrate the effectiveness of the DQN-based approach, achieving a coverage score of 0.80, a defense optimization (DO) of 0.77, and a risk mitigation effectiveness (RME) of 0.80. These findings underscore the ability of reinforcement learning to provide intelligent, adaptive defense mechanisms that enhance cybersecurity resilience. Furthermore, the study offers insights into how DQN can be integrated into real-world cybersecurity infrastructures to improve defense against sophisticated cyber threats.