Enhanced Intrusion Detection Systems via Feature Selection and Dimensionality Reduction Techniques

Due to the rapid development of modern networks, nowadays Intrusion Detection Systems (IDS) play a significant role in network security by detecting malicious activities or anomalies. However, the large dimensionality of network traffic data can undermine IDS efficacy, increasing processing complexity and perhaps reducing accuracy. This study explores these challenges by proposing a hybrid approach integrating Feature Selection (FS) and Dimensionality Reduction (DR) techniques. On the benchmark datasets NSL-KDD and KDDcup99, we compare various FS methods, such as filter-based, embedded-based, and wrapper-based, by selecting subsets of features to optimize IDS performance. Additionally, we explore Dimensionality Reduction techniques (DR) such as Principal Component Analysis (PCA) and Singular Values Decomposition (SVD), reducing the number of the features while maintaining accuracies of 88.91% and 70.61% on the KDDcup99 and NSL KDD test datasets and also with minimal runtime of 0.0513 and 0.0127 seconds respectively. Our findings show that hybrid FS and DR techniques improve IDS accuracy while drastically reducing computational overhead and increasing the efficiency and reliability of intrusion detection systems in real-world applications.