Communication Dans Un Congrès Année : 2026

Fed-SHAP-IDS: Federated SHAP-Based Intrusion Detection System for IoMT

Résumé

The use of the Internet of Medical Things (IoMT) has significantly increased in recent years, introducing new vulnerabilities and security challenges. Numerous attacks now target the healthcare ecosystem, exploiting weaknesses in connected medical devices. Traditional intrusion detection methods are increasingly insufficient to cope with the growing number of attacks and the emergence of zero-day threats. Artificial Intelligence (AI) has demonstrated strong potential in this context; however, centralized AI-based detection solutions expose sensitive medical data, create a single point of failure, and struggle with non-IID (non-independent and identically distributed) data across geographically dispersed devices. To overcome these limitations, we propose Fed-SHAP-IDS (Federated SHAP-based Intrusion Detection System), a novel federated learning framework specifically designed for IoMT environments. This system preserves data confidentiality by training models locally without sharing raw data, while effectively handling non-IID data distribution across heterogeneous clients. Furthermore, Fed-SHAP-IDS integrates the SHAP (Shapley Additive Explanations) method to capture local feature importance, enabling interpretable, explainable, and interoperable insights for each client. By combining federated learning with SHAP-driven feature aggregation, our approach enhances both privacy and transparency, offering a robust and trustworthy solution for IoMT intrusion detection.

Fichier non déposé

Dates et versions

hal-05552454 , version 1 (14-03-2026)

Identifiants

Citer

Mohammed Yacoubi, Omar Moussaoui, Cyril Drocourt. Fed-SHAP-IDS: Federated SHAP-Based Intrusion Detection System for IoMT. ICAART 2026 - 18th International Conference on Agents and Artificial Intelligence, Mar 2026, Marbella, Spain, Spain. pp.2683-2691, ⟨10.5220/0014315500004052⟩. ⟨hal-05552454⟩
86 Consultations
0 Téléchargements

Altmetric

Partager

  • More