A subspace-based few-shot intrusion detection system for the Internet of Things

• Proposes a subspace-based few-shot learning approach for IoT intrusion detection, effectively addressing the challenge of limited training samples. • Uses metric learning with subspace classifiers to detect unknown attack categories without the need for parameter optimization. • Constructs a few-shot IoT intrusion detection dataset based on CICIoT2023 and evaluates the method across 5-way 1-shot, 5-shot, and 10-shot settings. • Achieves high detection accuracy (up to 93.65%) for unknown categories, demonstrating robust generalization in sparse data scenarios.