Few-Shot Meta-Learning for Efficient Intrusion Detection in Contiki OS-based Wireless Sensor Networks

Wireless Sensor Network (WSN) is a more promising communication technique, but it is highly susceptible to attacks due to the lack of physical monitoring. This work focuses on enhancing Intrusion Detection Systems (IDS) using Contiki OS to simulate Denial-of-Service, Blackhole, and Selective Forwarding attacks. CC2538DK sensor nodes are programmed to inject these attacks and generate data for IDS training. To evaluate performance, the proposed work is executed on the generated and the WSN-DS publicly available IDS datasets. A key challenge in IDS development is the scarcity of labeled data for detecting new attacks. To address this problem, the method proposed a few-shot learning based IDS and applied a prototype-based neural network with a meta-learning strategy. Experimental results show that with a sample size of 25 or more per class, the model achieves an accuracy of approximately 96% while maintaining an accuracy of 93.2% with fewer than ten samples per class. This approach effectively detects new or previously unseen attacks.