A REPUTATION-AWARE FEDERATED LEARNING SYSTEM WITH DISTRIBUTED LEDGER INTEGRATION FOR SECURING MODEL CONTRIBUTIONS IN MULTI-AGENT SENSOR ENVIRONMENTS

Abstract

In the evolving landscape of edge intelligence and privacy-preserving AI, Federated Learning (FL) has emerged as a decentralized paradigm enabling collaborative model training without raw data sharing. However, FL remains vulnerable to poisoning attacks, unreliable client updates, and fairness issues. This study proposes a novel framework that integrates a reputation-aware mechanism with blockchain technology to ensure the reliability, transparency, and integrity of model contributions in multi-agent sensor environments. The system employs KMeans-based clustering to detect and filter low-quality or malicious updates and utilizes a lightweight blockchain ledger to immutably log verified contributions. Experimental results using the CMAPSS dataset demonstrate that the proposed framework significantly improves model robustness (R² = 0.8798) while preserving privacy and securing participation. The approach offers a scalable, secure, and trust-enhanced solution for real-world industrial IoT and autonomous systems.