Trust-Aware Adaptive Clustered Federated Learning for Energy–Latency Optimized WSNs

Abstract

Wireless sensor networks (WSNs) used in IoT applications have critical and intertwined challenges related to energy efficiency, communication delay, and security against malicious nodes. Existing federated learning (FL)-based solutions tackle these challenges individually, where energy-aware protocols neglect latency and trust-based schemes ignore communication delay. This paper proposes TASC-FL (Trust-Aware Self-Adaptive Clustered Federated Learning), a unified hierarchical FL framework that optimises energy consumption, communication delay and trust-based security simultaneously. TASC-FL introduces three main novelties: (i) a tri-factor node participation criterion based on the residual energy, trust score and communication delay; (ii) a reinforcement-based cluster head selection mechanism based on a composite reward function trade-off between the three factors; and (iii) a hybrid aggregation strategy based on weighting node contributions proportional to trust and energy with a penalty for delay. Simulation results via 1,000 rounds on a 100-node network show that TASC-FL achieves a network lifetime improvement of 95.3% over LEACH and 19.5% over DEEC; reduces per-round communication delay by 13.8% over SEAL-FL. converges to 88.2% accuracy, outperforming all baselines. preserves 44.5% accuracy under a 20% Byzantine attack versus 35.4% for SEAL-FL, a robustness gain of 9.1 percentage points. Our results make TASC-FL a scalable, secure and energy-efficient solution for next generation intelligent WSNs.