Towards Building Intrusion Detection Systems for Multivariate Time-Series Data

Abstract Recent network intrusion detection systems have employed machine learning and deep algorithms to defend against dynamically evolving attacks. While most previous studies focused on detecting attacks which can be determined based a single time instant, few paid attention subsequence outliers, require inspecting consecutive points in for detection. To address this issue, paper applies time-series anomaly method an unsupervised manner. end, we converted the UNSW-NB15 dataset into data. We carried out preliminary evaluation test performance of created as well obtained from sensors. analyze discuss results.