A Novel Dimension Reduction Based Method to Detect Network Intrusions

Abstract

The rapid growth of data and the increasing sophistication of cyber-attacks have made Intrusion Detection Systems (IDS) crucial in network security. However, traditional IDS techniques struggle with high false positive rates and inefficiencies in handling large-scale data. This study proposes a novel feature extraction method, Local Kernel-Principal Component Analysis (LKPCA), designed to enhance classification performance in IDS by incorporating class information into the dimensionality reduction process. The proposed method is evaluated using benchmark datasets: NSL-KDD, KDD, and CIC-IDS-2017. Experimental results show that LKPCA significantly improves classification accuracy and outperforming traditional methods like KPCA, PCA, and LSI. Additionally, LKPCA demonstrates superior efficiency by reducing training time compared to other feature extraction methods. This study also explores multi-class classifiers, where LKPCA further enhances attack detection performance. The findings suggest that LKPCA, when combined with classifiers such as Support Vector Machines (SVM) and Decision Trees (DT), provides a powerful tool for improving the effectiveness and efficiency of IDS, making it a promising solution for real-time network security applications.