Answers are grounded in this paper and cite their sources.
Intrusion detection in distributed IoT environments is constrained by data-locality requirements, heterogeneous traffic distributions, changing attack patterns, and limited model interpretability. This work presents Fed-XAI-HMAIDS, an empirically evaluated federated extension of the Hierarchical Multi-Agent Intrusion Detection System (HMAIDS). The framework integrates a global One-Class Support Vector Machine (OCSVM) for anomaly screening, federated one-dimensional Convolutional Neural Network (1D-CNN) classification, sample-size-weighted FedAvg aggregation, continual learning and SHAP-based feature attribution within the HMAIDS detection–coordination–explanation architecture. Experiments on CICIoT2023 evaluate anomaly detection, federated classification, different non-IID client distributions, alternative federated configurations, repeated random-seed runs, and replay-based continual learning. The global OCSVM achieves an 83.54% detection rate with a 5.08% false alarm rate, while the three best federated configurations achieve mean accuracies ranging from 71–73% across ten repeated runs. The full pipeline achieves 71.8% accuracy in the representative training-paradigm comparison, compared with 73.0% for the corresponding FedAvg CNN baseline. SHAP analysis identifies network protocol-, flow-, and transport-related features that contribute to attack classification. The results provide empirical evidence for the feasibility of integrating anomaly screening, federated classification, continual learning, and post-hoc explainability within the HMAIDS architecture while retaining raw network traffic at the client level. The findings are limited to the evaluated dataset, four-client configuration, partition strategy, and threat model, and do not establish detection of previously unseen or zero-day attack classes.
No takes yet. Share an insight, caveat, or question.
Suaifan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: