Intrusion Detection using Spatial-Temporal features based on Riemannian Manifold

Abstract

Network traffic data is combination of different data bytes packet under different network protocols. These traffic packets have complex time-varying non-linear relationships. Existing state of the art method rise up to this challenge by fusing features into multiple subset based on correlations and using hybrid classification techniques that extract spatial and temporal characteristics. This offen requires high computational cost and manual support that limit them for real-time processing of network traffic. To address this, we propose a new novel feature extraction method based on covariance matrices that extracts spatio-temporal characteristics of network traffic data for detecting malicious network traffic behaviour. The covariance matrices in our proposed method not just naturally encodes the mutual relationships between different network traffic values but also have well defined geometry that falls in Riemannian manifold. Riemannian manifold is embedded with distance metrices that facilitates extracting descriminative features for detecting malicious network traffic. We evaluated our model on an NSL-KDD and UNSW-NB15 datasets and shown our proposed method significantly outperforms the conventional method and other existing studies on the dataset.