Enhancing detection accuracy of cyber attacks through dimensionality reduction

Author ORCID Identifier

0000-0002-9956-4003 : Ehsan Hallaji

0000-0002-4330-3656 : Roozbeh Razavi-Far

Document Type

Conference Proceeding

Publication Date

11-2020

Publication Title

Proceedings of the 30th European Safety and Reliability Conference and the 15th Probabilistic Safety Assessment and Management Conference

First Page

1346

Keywords

Intrusion detection, cyber-physical systems, gas pipeline, dimensionality reduction, machine learning, SCADA.

Last Page

1351

Abstract

The importance of cyber-security has led to long-standing endeavors dedicated to the design of intrusion detection systems (IDS). Nevertheless, the performance of these data-driven techniques is highly dependent on data quality. Incorporating dimensionality reduction techniques into a hybrid intrusion detection system, we aim to study the effect of dimensionality reduction on the performance of intrusion detection. By this mean, the efficiency of the intrusion detection systems is increased by processing a smaller feature space. Moreover, the reduced feature space also increases the detection accuracy, as redundant and meaningless features are removed in the new feature space. Furthermore, the intrinsic structure of the data is improved, that is different states of the system become more discriminant after dimensionality reduction. For this mean, various state-of-the-art dimensionality reduction techniques are selected. Then, a simulation is performed on a Supervisory Control and Data Acquisition (SCADA) system, which resembles a gas pipeline control system introduced by Morris et al. (2011). A comparative study is then performed to suggest the best dimensionality reduction algorithm in these experiments. The experiments indicate the general improvement of detection accuracy when dimensionality reduction techniques are combined with the IDS in terms of accuracy and standard deviation.

DOI

10.3850/978-981-14-8593-0_4593-cd

ISBN

978-981148593-0

Share

COinS