A Study on Free-Rider Detection Mechanism for a Fair Federated Learning-Based Intrusion Detection System

Federated Learning (FL) has been increasingly adopted for building privacy-preserving Intrusion Detection Systems (IDSs) in distributed and IoT environments. However, free-rider attacks—where participants benefit from the global model without contributing genuine updates—undermine both fairness and detection performance. This paper introduces a blockchain-enabled decentralized FL framework that integrates a Deep Autoencoding Gaussian Mixture Model (DAGMM) to detect and mitigate free-riders. The proposed system leverages smart contracts and decentralized storage to ensure model integrity while applying anomaly-based client validation to penalize selfish and random weight submissions. Extensive experiments on two IoT intrusion detection datasets, MQTTset and NF-ToN-IoT-v2, demonstrate that our approach achieves stable detection accuracy above 96% and improves F1-scores by up to 17% in the presence of 20% free-riders, compared to baseline FL-IDS. These results highlight the necessity of proactive free-rider detection for secure and equitable collaborative intrusion detection in decentralized networks.