Efficient Dimensionality Reduction in Intrusion Detection Systems via Weighted Power Mean for PCA and LDA

The rapid growth of digital networks has highlighted the critical need for effective Intrusion Detection Systems (IDS). However, the increasing data volume presents challenges for conventional IDS, leading to more false alarms and reduced accuracy. In this study, we propose an innovative approach that combines Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA), emphasizing robustness. To mitigate outlier susceptibility inherent in PCA and LDA, we introduce the generalized sample mean. Our experiments, conducted on the UNSW-NB15 dataset, demonstrate the superior performance and resilience of our approach compared to traditional methods.