EXPLAINABLE ARTIFICIAL INTELLIGENCE MODELS IN INTRUSION DETECTION SYSTEMS


Sağıroğlu Ş., Sağıroğlu Ş.

ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE, vol.1, no.1, pp.1-10, 2024 (SCI-Expanded)

  • Publication Type: Article / Article
  • Volume: 1 Issue: 1
  • Publication Date: 2024
  • Journal Name: ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Aerospace Database, Applied Science & Technology Source, Communication Abstracts, Compendex, Computer & Applied Sciences, INSPEC, Metadex, Civil Engineering Abstracts
  • Page Numbers: pp.1-10
  • Gazi University Affiliated: Yes

Abstract

With the increasing adoption of artificial intelligence (AI) technologies, popular, incomprehensible, complex, and opaque machine learning (ML) models, especially neural network models, are becoming increasingly difficult to understand. This situation worsens the problem in an area such as cyber security and makes it important. Trusting a system that cannot explain the reason for important decisions and leaving it alone brings with it many concerns and can sometimes involve obvious dangers. In particular, the complexity of AI models is increasing, resulting in black box models that cannot be easily examined, verified, or tested. To overcome this problem, Explainable AI (XAI) proposes approaches that can be a solution to this problem with more interpretable, explainable, and understandable AI models and the resulting outputs. XAI, a holistic approach to the solution, uses various methods to understand, comprehend, and interpret AI models and even show what data or data regions the decision was made based on. XAI provides frameworks that help understand and explain the predictions of AI algorithms, bridging between real intelligence and AI. Although the concept of XAI and the developed models have attracted great attention recently and are used more intensively in some areas, they are not yet used sufficiently in Intrusion Detection Systems (IDSs). It is important for IDSs, which are an important solution in detecting data and attacks, to have high performance and to explain decisions or learn their justifications. In addition, in order to develop some examples and carry out studies that can guide future research, there is a need to apply XAI methods in IDSs, explain the decisions taken and outputs, analyze the results obtained and convert them into explainable forms or formats. In this study, XAI has been examined in general and evaluated from different perspectives, and XAIs and their applications in IDSs have been examined in detail. A detailed explanation of definitions and terminology in the evolving field of XAI has been put together; Opportunities, challenges, and areas where further research is needed in the field were examined; Approaches and latest developments, tools, and technologies for developing XAI applications, as well as the issues that need to be done for their implementation in artificial intelligence-based IDSs and the risks encountered are summarized