Gaussian process latent variable model for dimensionality reduction in intrusion detection

As the interest in internet grows the issue of security and defending crucial systems against cyber-attacks become more important. Intrusion detection systems (IDSs) are considered as security solutions that can protect our information systems against threats. However, the datasets on which IDSs are applied are high dimensional which is the reason of low performance of these systems from the view of time and space complexity. We have proposed an approach based on dimensionality reduction to reduce this complexity. In our approach, the GPLVM is used as the main method because of its inherent attributes such as being nonlinear and nonparametric. The effectiveness of our approach is demonstrated with experiments on a benchmark dataset (NSL-KDD). To visualize the proposed approach we reduced the dataset to 2-dimensional one and show that even in this reduced dataset, our approach is promising in terms of accuracy for real-world intrusion detection.