A Meta-Learning Approach for Few-Shot Network Intrusion Detection Using Depthwise Separable Convolution

Abstract

As cyberattacks become more frequent and sophisticated, network intrusion detection systems (IDS) play a critical role in safeguarding networks. However, traditional IDS models face challenges in detecting new, unseen attacks and typically require large volumes of labeled data for effective training. To address these issues, we propose a novel intrusion detection model based on meta-learning, integrating depthwise separable convolution (DSC). This model leverages few-shot learning to detect rare and emerging attack types with minimal labeled data. By using meta-learning, our model can rapidly adapt to new tasks, offering greater flexibility and scalability in various network scenarios. Experimental results on the CIC-DDoS2019 and CIC-IDS2017 datasets demonstrate that our model achieves competitive accuracy compared to state-of-the-art methods, even with fewer training samples. It also shows superior performance in terms of both detection accuracy and training efficiency, while being more resource-efficient, making it suitable for deployment in resource-constrained environments. In conclusion, our model offers a promising solution for network intrusion detection, enhancing the ability to detect new and emerging threats while ensuring computational efficiency for real-world applications.