Latency and trust constrained fog node selection using deep reinforcement learning

Automated healthcare IoT systems demand secure, low-latency, and energy-efficient computation—capabilities well-supported by fog computing. Effective selection of fog nodes is critical for maximizing the performance of fog-based IoT platforms. This paper introduces a Secure Proximal Policy Optimization (Secure PPO) algorithm for trust-aware fog node selection, considering latency, energy consumption, processing power, and a trust flag for each node. Secure PPO enforces a trust constraint while optimizing latency and energy via PPO's clipped objective, ensuring stable and reliable learning. Simulation results demonstrate that Secure PPO achieves substantial improvements over A2C and Deep Q-Networks (DQN). Specifically, Secure PPO reduces inference latency by 24.36 and 37.57%, lowers convergence time by 55.56 and 66.67%, and decreases energy consumption by 11.90 and 20.04% compared to A2C and DQN, respectively. Additionally, Secure PPO improves accuracy by 9.42 and 18.88% over A2C and DQN. The framework maintains sub-millisecond inference time and ensures secure, reliable fog-based execution of automated healthcare tasks, substantially enhancing patient safety and operational efficiency within healthcare IoT environments.