Secure Routing in Ad Hoc Networks through Federated GNN Trust Aggregation and Topology-aware Flow Analysis

Trust-aware routing in Mobile Ad Hoc Networks (MANETs) is vital for secure communication under dynamic and adversarial conditions.Centralized trust computation improves reliability but introduces scalability constraints and privacy risk during model training.This work presents TrustFed-GNN, a decentralized routing framework that couples Graph Neural Network (GNN) based trust modeling with Federated Learning (FL) to enable collaborative training without exposing raw traffic data.The graph encoder propagates trust over topology-aware neighborhoods, while the federated protocol aggregates local updates into a shared model that remains privacy preserving.Experiments on the CIC-IoT-2023 dataset, comprising 3.12 million flows with 46 attributes spanning benign and six attack classes, show an accuracy of 98.9%, an F1-score of 96.8%, and a packet delivery ratio of 95.3% under Sybil and Blackhole stress tests.Relative to AODV-Trust, the approach reduces the false positive rate by 34.3% and energy usage by 28.8%.These findings indicate that topology-aware learning combined with privacy-preserving coordination provides effective, resource-conscious, and attack-resilient routing for Internet of Things (IoT) enabled ad hoc deployments.