IDS for Industrial Applications: A Federated Learning Approach with Active Personalization.

Area of Science:

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

Background:

• The widespread adoption of Internet of Things (IoT) devices, particularly in industrial sectors for critical infrastructure monitoring and control, has significantly expanded the attack surface.
• Traditional machine learning methods for securing IoT can compromise sensitive data.
• There is a need for privacy-preserving and adaptive AI solutions for IoT cybersecurity.

Purpose of the Study:

• To develop a network flow-based Intrusion Detection System (IDS) for protecting industrial IoT critical infrastructures.
• To leverage federated learning for private model training and active learning for adaptive global model personalization.

Main Methods:

• Implementation of a novel Intrusion Detection System (IDS) using a combination of federated learning and active learning techniques.
• Federated learning enables private, collaborative model training across distributed participants.
• Active learning is employed for semi-supervised, efficient global model adaptation to individual participant traffic patterns.

Main Results:

• Experimental results demonstrate that globally trained models, when locally personalized via active learning, significantly improve performance for each participant.
• An accuracy increase of up to 7.07% was achieved with only 10 active learning queries.
• The proposed approach enhances the effectiveness of AI-driven intrusion detection in IoT environments.

Conclusions:

• The pairing of federated learning and active learning offers a robust and privacy-preserving solution for securing critical IoT infrastructures.
• Local personalization of global models through active learning is highly effective in improving IDS performance.
• This AI-driven cybersecurity approach addresses the challenges posed by the expanding attack surface in industrial IoT.