Meta-IDS: A Multi-Stage Deep Intrusion Detection System with Optimal CPU Usage
Nadia Niknami, Vahid Mahzoon, Jie Wu · 2024
The exponential growth in network traffic coupled with the intricacies of neural network methodologies has presented formidable challenges for conventional single-machine architecture Network Intrusion Detection Systems (NIDS). In this paper, we propose a novel approach to address these challenges drawing inspiration from meta-computing principles. Meta-computing, characterized by dynamic resource allocation, offers a promising solution for optimizing NIDS performance in the face of escalating network traffic. We introduce a hierarchical NIDS that employs varying levels of model complexity to efficiently process different scales of network traffic. In designing the proposed lightweight IDS, our efforts focus on selecting the optimal set of features for each level. By integrating dynamic resource allocation techniques, our Meta-IDS is able to adapt proactively to fluctuations in network activity, thus mitigating the risk of resource depletion while preserving its efficacy in detecting intrusions. Through experiments conducted on a benchmark dataset, we demonstrate the effectiveness of our proposed Meta-IDS in enhancing network intrusion detection in high-performance environments. Our approach not only ensures efficient resource management but also enhances the accuracy and timeliness of intrusion detection, thereby providing a robust solution for modern network security challenges.