Neural Network-Based Task Offloading in MEC: Performance Evaluation and Hybrid Model Optimization

Mobile Edge Computing (MEC) has emerged as a promising approach to reduce the latency of computational task offloading to nearby edge servers, thereby enhancing user experience. Artificial Intelligent (AI) has been shown to improve task offloading in terms of efficiency, latency reduction, energy consumption, and resource utilization. However, selecting the most effective approach for optimal decision-making in task offloading remains a significant challenge. In this paper, we propose a comprehensive evaluation framework to identify the most efficient Neural Network Model (NNM) for optimal task offloading decisions in MEC. We conducted a detailed comparative analysis of various individual NNMs and their hybrid models, evaluating their performance and computational efficiency on multiple metrics. The simulation results demonstrate that the Gated Recurrent Unit (GRU) model consistently achieves outstanding results, it has achieved an average accuracy of 95.10%. Similarly, the hybrid model of Long Short-Term Memory (LSTM) and GRU also delivers excellent performance by achieving an average accuracy of 94.88%, in calculating the minimum cost for all possible combinations of task offloading in MEC. This study provides a valuable guide for optimize offloading in MEC systems and highlights the potential of leveraging advanced NNMs for efficient decision-making of offloading in MEC.