Agents Incorporating XRL-Based Method for Threat Detection in Mobile Edge Networks for Consumer Electronics
Muhammad Yousaf Saeed, Jingsha He, Nafei Zhu, Muhammad Farhan, Mrim M. Alnfiai, Juan Fang · IEEE Transactions on Consumer Electronics · 2025
Real-time threat detection in mobile edge networks for consumer electronics based on the new Explainable AI (XAI) technique for reinforcement learning framework is proposed. The approach leverages Advantage Actor-Critic (A2C), Proximal Policy Optimization (PPO), and Hindsight Experience Replay (HER) to improve learning efficiency in dynamic threat detection scenarios. Continuous interaction with the network environment allows the RL agent to adapt to the appearance of new threats that can raise threat detection precision and resilience. We use SHapley Additive exPlanations (SHAP) to build additively interpretable models, offering practical insights to network operators while enhancing transparency and confidence in the decision process. Extensive empirical evaluations with benchmark datasets show that our method substantially outperforms traditional ones by up to 15% in accuracy, with an extra 30% false positives reduction. It realizes 99.97%, 99.98%, and 99.95% on the respective proposed A2C, PPO, and HER models for networks identified under severe attack scenarios, outperforming other RL algorithms like SARSA, with 99.69%, and Q-learning, with 98.53%. These results confirm the very competitive performance of the proposed RL models in handling complex classification problems compared to classical machine learning models such as Random Forest and Multi-layer Perceptron. The integration with lightweight XAI helps ensure RL agents’ decisions are correct and interpretable, solving yet another fundamental challenge in deploying AI-based security systems within sensitive environments for consumer electronics.