Efficient Network Intrusion Detection using Grassmann Manifold

Recently, the increasing complexity and volume of network traffic have posed significant challenges for conventional Network Intrusion Detection Systems (NIDS). Traditional NIDS often rely on predefined patterns and labeled datasets, making them inadequate for handling dynamic, high-speed, and high-dimensional network traffic. To address this issue, recent studies have investigated the integration of machine learning into NIDS. However, existing approaches are often dependent on supervised learning models that are inefficient and heavily reliant on labeled datasets. Moreover, these models struggle to process high-dimensional data and fail to generalize to unseen attacks. To this extent, this paper explores an alternative approach by introducing a framework that leverages Grassmann manifolds to overcome these limitations. Grassmann manifolds, which represent the set of all possible subspaces of a given dimension, provide a robust geometric foundation for projecting network traffic data. Leveraging this structure, the proposed method effectively captures abnormal patterns and structural information, enabling precise identification of anomalous behaviors. Comprehensive experiments using the NSL-KDD dataset demonstrate that the proposed GrassmannPCA approach not only improves detection accuracy, precision, and F1-score, but also achieves lower false alarm rates and reduced computational cost. The proposed GrassmannPCA framework outperformed EuclideanPCA, achieving an accuracy of 82.97%, a precision of 88.32%, an F1-score of 84.37%, and a significantly lower false alarm rate of 14.12%, while also reducing training time by 1.12%.