SVDCNet: SVD-Driven Convolutional Neural Network for Advanced Multi-Label Network Intrusion Detection

Network intrusion detection is a crucial defense technique in the ever-evolving realm of cybersecurity, where the proliferation of sophisticated cyber-attacks demands highly accurate and resilient detection models. The growing complexity and variety of these attacks necessitate advanced methods capable of adapting to new and unknown threats. However, existing approaches often fail to achieve a balance between accuracy and robustness, leading to performance degradation in diverse network environments. This inadequacy underscores the need for innovative solutions that can maintain high detection performance across varying conditions and attack scenarios. To address these challenges, SVDCNet, an innovative Convolutional Neural Network (CNN) model enhanced with Singular Value Decomposition (SVD), is proposed to optimize both accuracy and robustness in detecting multiple types of network intrusions. SVDCNet's architecture integrates SVD to enhance model stability and generalization, enabling consistent performance across various datasets. Experimental evaluations on the CICIDS 2017, CICIDS 2018, and UNSW-NB15 datasets demonstrate SVDCNet's superior capabilities, achieving high accuracy rates of 97.63%, 99.98%, and 97.21% respectively. These results underscore SVDCNet’s effectiveness in providing a robust and accurate solution for multi-label network intrusion detection, surpassing traditional models that prioritize accuracy at the expense of robustness.