Performance Analysis of AI-based Intelligent Computation Offloading in Edge Computing Using Reinforcement Learning, ACO and GA

Abstract

Computation offloading in Edge Computing (EC) plays a vital role in reducing execution latency and improving quality of service for resource-constrained devices. This paper presents a comparative performance analysis of three intelligent task offloading strategies, namely Reinforcement Learning (RL), Ant Colony Optimization (ACO), and Genetic Algorithm (GA), with the objective of minimizing overall system latency. A simulation-based system model is developed to evaluate the effectiveness of these approaches under dynamic network conditions. The experimental results demonstrate that the RL-based method achieves superior performance, attaining an average latency of approximately 42 ms, which is significantly lower than ACO and GA. This improvement highlights the adaptive learning capability and faster convergence of RL in making optimal offloading decisions. The findings indicate that RL-based offloading is more suitable for dynamic EC environments, and the proposed framework can be extended to incorporate additional metrics such as energy efficiency and throughput in future work.