1.
Sushma Malik
– Assistant Professor, Maharaja Surajmal Institute, New Delhi, India.
2.
Anamika Rana
– Associate Professor, Maharaja Surajmal Institute, New Delhi, India.
Abstract
Cyber Threat Intelligence (CTI) is a key component of cutting-edge security that empowers businesses and organisations to identify, prevent and even predict new cyber threats. Rule-based systems, on the other hand tend to be narrowly defined and are not flexible and do not capture Advanced Threat Detection for complex or even multi-layer or zero-day attacks. Novel Machine Learning (ML) techniques have also been considered a promising alternative, but existing studies rarely aim to perform a comprehensive comparative analysis of these techniques in the specific context of CTI frameworks. This study provides an opportunity to evaluate a few ML Algorithms (Support Vector Machines, Random Forest, K-Means Clustering and Neural Networks) to enhance CTI. The study performs a systematic literature review and comparative analysis through a selection of recent papers/datasets on their performance in the three most important domains: perceived intrusions, malware classification and phishing detection for several services. It is observed that ensemble and deep learning models provide better accuracy, scalability on evolving cyber systems, as well as generalisation to new data than the traditional ML models, while unsupervised techniques are useful for detecting zero-day threats. In this article, we reassess major challenges of ML-based CTI systems namely data imbalance, high computational complexity and adversarial attack vulnerability which may seriously undermine the system reliability robustness after deployment in practical scenarios. It provides insights and addresses a seldom-discussed gap in the literature by conducting a critical review, evaluating current ML methods for CTI, identifying major areas requiring further research and offering an integrated theoretical perspective on how ML can be used to provide automated support systems for CTI practices. Results show that it is important to have more powerful, scalable and adaptive ML models that will allow smart, proactive, resilient cybersecurity postures against the continuous evolution of cyber threats.
Keywords Adversarial attacks, Big data analytics, Cybersecurity, Cyber Threat Intelligence (CTI), Explainable AI (XAI), Federated learning, Machine learning (ML), Malware analysis, Network security, Phishing detection, Real-time threat analysis, Threat detection.