A Hybrid Genetic-Reinforcement Learning Framework for Edge-Enabled Social IoT

Abstract

The Social Internet of Things (SIoT) enables socially connected devices to deliver context-aware and personalized services. However, traditional SIoT systems are typically cloud-based, leading to high latency, high bandwidth consumption, privacy concerns, and limited adaptability to user preferences. To address these issues, this study introduces a novel Coevolutionary Hybrid Intelligence (CHI) system that combines Genetic Algorithms (GAs), Reinforcement Learning (RL), and Edge AI to enable adaptive and collaborative SIoT systems. The proposed framework is built using a distributed edge architecture that continuously updates device policies and user preference models by incorporating: on-device intelligence, local inference, user feedback, and a coevolutionary optimization engine. Experimental testing in a simulated smart-home environment demonstrates that CHI can deliver a response time of 123 ms, a bandwidth utilization of 84 MB/day, an energy consumption of 8.8 W, and a user satisfaction score of 4.7/5. The proposed model is compared to cloud-centric and edge-only baselines, and the results indicate better responsiveness and lower communication overhead. The results show that CHI is an effective and scalable solution for privacy-preserving and low-latency personalized SIoT applications.