Review Article

Artificial Intelligence and Machine Learning in Cyber Threat Detection for Modern Organizations: A Systematic Review of Challenges and Solutions

Authors

• • Dr. Abdul Wahid Khan wahidkn@gmail.com Email Dr. Abdul Wahid Khan. (compose email, opens in email app.)
  • Department of Computer Science, College of Engineering and Information Technology
  • Onaizah Colleges
• • Bassam Saleh F Alghashim
• • Meshari Khalid N Alharb
• • Tamim Sulaiman A Alghomiz
• • Faisal Fahad M Alshrahrani
• • Faisal Saud F Almsaad

Abstract

The ability to adapt to threats using AI and ML techniques has become a major factor in the fight against cyber threats, offering a data-driven, adaptive approach to threat detection that transcends signature-based methods. With the growing surge of research and development on AI/ML-based intrusion detection, however, there has been disarray in the perception of the challenges which organizations are facing and the solutions available to them to tackle the challenges. This paper is based on a systematic literature review (SLR) carried out following the Kitchenham SLR protocol and a PICOC search strategy, and the analysis of 45 peer-reviewed and industry-reported studies related to the application of AI/ML for cyber threat detection. In total, the corpus identified 271 different challenges and 252 solutions, which were grouped into seventeen recurring categories of challenges. The biggest challenges to successful adoption were zero day and novel attack detection (36), scalability and big data (34), high false positive/negative rates (33), computational and resource constraints (32), and dataset imbalance (31), followed by outdated datasets, privacy and regulatory compliance, explainability, and high dimensional feature spaces. The comparison on the continental level also identified the scalability and IoT-based detection as the main challenge in Asia, privacy and explainability as the primary challenge in Europe, adversarial robustness and enterprise security as the main challenge in North America and foundational infrastructure and capacity limitations as the main challenge in the Middle East/Africa. The findings indicate a trajectory towards more hybrid, adaptive, and privacy-conscious approaches to AI security that integrate ensemble and deep learning techniques, federated learning, explainable AI, and ongoing training pipelines. This research work highlights the severity of challenges faced by the future organisations in implementation of AI/ML for cybersecurity intrusion detection.

Keywords:

Artificial Intelligence, Machine Learning, Systematic Literature Review, Cybersecurity Challenges, Solutions, Modern Organisations

How to Cite: