FD-IDS: Federated Learning with Knowledge Distillation for Intrusion Detection in Non-IID IoT Environments.

Area of Science:

• Cybersecurity
• Machine Learning
• Internet of Things (IoT)

Background:

• Traditional centralized learning struggles with sensitive, Non-IID IoT data.
• Ensuring data privacy and handling distributed heterogeneity are critical challenges in IoT security.
• Existing intrusion detection systems (IDSs) face limitations with the unique characteristics of IoT data.

Purpose of the Study:

• To propose a novel IDS framework, FD-IDS, leveraging federated learning and knowledge distillation.
• To address data privacy concerns and the complexities of Non-IID data in IoT environments.
• To improve the efficiency and effectiveness of intrusion detection in IoT networks.

Main Methods:

• Developed FD-IDS framework integrating federated learning and knowledge distillation (KD).
• Employed mutual information for efficient feature selection.
• Combined a proximal term with KD to mitigate model drift in Non-IID scenarios.

Main Results:

• FD-IDS demonstrated promising detection performance on Edge-IIoT and N-BaIoT datasets.
• The framework effectively addressed data privacy and distributed heterogeneity issues.
• Experimental results validated the system's ability to alleviate model drift.

Conclusions:

• FD-IDS offers a robust solution for secure and efficient intrusion detection in IoT.
• The integration of federated learning and KD proves effective for handling Non-IID IoT data.
• The proposed methods enhance training efficiency and detection accuracy in complex network environments.