Application-Aware Intrusion Detection: A Systematic Literature Review, Implications for Automotive Systems, and Applicability of AutoML

• Main Authors: David Schubert, Hendrik Eikerling , Jörg Holtmann
• Format: Article
• Language: English
• Published: Frontiers Media S.A. 2021-08-01
• Series: Frontiers in Computer Science
• Subjects: intrusion detection; security engineering; survey; AutoML; automotive
• Online Access: https://www.frontiersin.org/articles/10.3389/fcomp.2021.567873/full

Modern and flexible application-level software platforms increase the attack surface of connected vehicles and thereby require automotive engineers to adopt additional security control techniques. These techniques encompass host-based intrusion detection systems (HIDSs) that detect suspicious activities in application contexts. Such application-aware HIDSs originate in information and communications technology systems and have a great potential to deal with the flexible nature of application-level software platforms. However, the elementary characteristics of known application-aware HIDS approaches and thereby the implications for their transfer to the automotive sector are unclear. In previous work, we presented a systematic literature review (SLR) covering the state of the art of application-aware HIDS approaches. We synthesized our findings by means of a fine-grained classification for each approach specified through a feature model and corresponding variant models. These models represent the approaches’ elementary characteristics. Furthermore, we summarized key findings and inferred implications for the transfer of application-aware HIDSs to the automotive sector. In this article, we extend the previous work by several aspects. We adjust the quality evaluation process within the SLR to be able to consider high quality conference publications, which results in an extended final pool of publications. For supporting HIDS developers on the task of configuring HIDS analysis techniques based on machine learning, we report on initial results on the applicability of AutoML. Furthermore, we present lessons learned regarding the application of the feature and variant model approach for SLRs. Finally, we more thoroughly describe the SLR study design.