Abstract

Fog computing has emerged as a promising paradigm for
supporting latency-sensitive healthcare services by bringing computation closer to Internet of Things (IoT) devices. Efficient task offloading in such systems is inherently a multi-objective optimization problem, where latency, energy consumption, throughput, task completion rate,
and fairness must be jointly optimized. This paper presents a comparative exploration of three distinct optimization approaches for healthcare IoT–fog task offloading: Reinforcement Learning (RL), Genetic Algorithm
(GA), and Mixed-Integer Linear Programming (MILP). Each
method represents a different paradigm—learning-based, evolutionary, and deterministic mathematical optimization. A utility function is formulated to capture the trade-offs across multiple objectives, and the algorithms are evaluated under varying fog node configurations. Simulation results demonstrate that RL consistently achieves lower latency and
energy consumption while balancing fairness and throughput, GA provides stronger performance in task completion and resource utilization, and MILP offers moderate but less adaptive results. The comparative analysis highlights the trade-offs between exploration capability, adaptability,
and computational cost, establishing RL as the most effective
and scalable solution for dynamic healthcare IoT–fog environments.