Enhancing Cybersecurity in Healthcare IoT Systems Using Reinforcement Learning

Due to the diversity of devices, vast attack surfaces, and the critical nature of healthcare data, securing healthcare networks within the Internet of Things (IoT) presents a challenging task. This study addresses the heightened cybersecurity risks associated with the growth of IoT devices in healthcare by developing a custom Reinforcement Learning (RL) cybersecurity IoT environment for healthcare. For this purpose, we operate to adopt Proximal Policy Optimization (PPO), and Advantage Actor-Critic (A2C) RL models to simulate attack and defense strategies, assess network resilience, and identify effective cybersecurity practices. The proposed approach employs Microsoft CyberBattleSim, a simulation environment for reinforcement learning, along with Gymnasium, to construct a healthcare IoT environment incorporating diverse devices, vulnerabilities, and security protocols. The experiments showcased the capability of the proposed models to adjust policies for addressing complex cyber threats, thus revealing essential defense mechanisms and prevalent vulnerabilities within the network. Our results showed that A2C performed slightly better than PPO in terms of average and maximum rewards, possibly due to its on-policy learning mechanism enabling faster adaptation to environmental changes. However, PPO exhibited a more stable learning curve, albeit with slower convergence.