A Systematic Literature Review on Cyber Threat Detection Using Machine Learning Techniques: Cyber Threat, Algorithms, and Regional Perspectives

The rapid escalation in the scale and sophistication of cyber threats has intensified interest in leveraging machine learning (ML) for proactive cybersecurity defense. This systematic literature review (SLR), guided by the PRISMA methodology, synthesizes studies published between 2016 and 2024 to identify machine learning algorithms demonstrating the highest effectiveness in predicting cyber threats. Initially, 350 articles were retrieved from three academic databases (Google Scholar, IEEE Xplore, and ScienceDirect). After applying rigorous inclusion and exclusion criteria, 74 articles were selected for detailed analysis. This review identified five dominant categories of cyber threats: phishing, ransomware, denial-of-service (DoS), cloud-based intrusions, and supply chain attacks. Additionally, the review assessed the performance of various ML algorithms discussed in the literature. Random Forest emerged as the most frequently employed and consistently effective classification algorithm, followed by Decision Trees, Support Vector Machines (SVM), and Naive Bayes classifiers. While findings indicate significant global progress, the review emphasizes a notable research gap in underrepresented regions such as the South Pacific, where ML applications in cybersecurity remain limited. The outcomes of this review provide a foundation for future research aimed at developing adaptive, ML-driven cyber defense systems tailored to the specific needs of the South Pacific and similar regions globally. Overall, this study contributes to the field by highlighting critical intersections between machine learning and cybersecurity.