Federated Learning for IoMT Applications: A Standardization and Benchmarking Framework of Intrusion Detection Systems

Area of Science:

• Cybersecurity in healthcare
• Machine learning applications
• Federated learning frameworks

Background:

• The Internet of Medical Things (IoMT) requires robust intrusion detection systems (IDSs).
• Evaluating machine learning (ML)-based IDSs in federated learning (FL) environments presents standardization challenges.
• Multicriteria decision-making (MCDM) offers a structured approach to complex evaluation problems.

Purpose of the Study:

• To develop an MCDM framework for standardizing and benchmarking ML-based IDSs within IoMT applications using FL.
• To establish consensus on evaluation criteria for ML-based IDSs in IoMT.
• To identify the optimal ML-based IDS for IoMT security and performance.

Main Methods:

• Standardization of ML-IDS evaluation criteria using the fuzzy Delphi method (FDM).
• Formulation of an evaluation decision matrix (DM) using a dataset of 125,973 records with 41 features.
• Integration of MCDM methods, including Borda voting and VIKOR, for criteria weighting, benchmarking, and optimal IDS selection.

Main Results:

• Consensus reached on 17 out of 20 evaluation criteria (14 security, 3 performance) via FDM.
• CPU time identified as the most critical criterion, while area under the curve received the lowest weight.
• VIKOR group ranking indicated BayesNet as the superior classifier, with SVM performing last.

Conclusions:

• The proposed MCDM framework effectively standardizes and benchmarks ML-based IDSs for IoMT environments.
• BayesNet emerges as the most suitable ML-based IDS for IoMT applications based on the established criteria.
• The study provides a systematic approach for evaluating and selecting optimal IDSs in critical healthcare networks.