MA-RL and WOA-Based Secure Routing: A Federated Learning Approach for Trust Evaluation and DoS Attack Mitigation in IoT Networks

Abstract

The growing size and sophistication of the Internet of Things (IoT) has brought up the issue of network security and routing efficiency, especially in unfriendly settings where Denial-of-Service (DoS) attacks are frtaent. Conventional trust-based routing algorithms are not very flexible and are not capable of adequately addressing these dynamic threats. This paper puts forward a well-constructed smart and secure routing architecture that incorporates Multi-Agent Reinforcement Learning (MA-RL), Whale Optimization Algorithm (WOA), and Federated Learning (FL) to carry out more sophisticated trust scoring and DoS attack early warning. The framework is designed in three central steps, namely: (1) dynamic trust by MA-RL on the basis of metrics such as Packet Forwarding Ratio (PFR), rewards/penalties, and Received Signal Strength Indicator (RSSI); (2) privacy-preserving DoS detection via FL-based decentralized anomaly detection; and (3) optimized secure path selection with the use of WOA, where there is trust, energy efficiency, and a low latency. Moreover, a predictive time-series model predicts degradation of trust, which enables malicious nodes to be isolated in advance. The experimental assessment of the simulated IoT settings during the DoS attacks proves considerable positive changes in the detection rate, routing speed, and network stability. The proposed model has a high potential of securing the next-generation IoT networks as it has an adaptive learning capacity which is scalable.