Optimizing Cybersecurity Threat Detection Using Machine Learning: A Comparative Study of Supervised and Unsupervised Approaches

Abstract

As cyber-attacks become more sophisticated, conventional rule-based security systems are no longer adequate for effective threat detection in a timely manner. In this study, the effectiveness of machine learning (ML) algorithms to detect and respond to cybersecurity threats is compared using supervised and unsupervised learning approaches. Models are trained on benchmarking datasets like CICIDS2017 and NSL-KDD to analyze detection rate, false-positive rates, and computational complexity. Results show that supervised models such as Random Forest and Support Vector Machine are more accurate compared to unsupervised models, but the clustering-based methods have strong zero-day attack detection anomaly. These results validate a hybrid model that incorporates the strengths of both learning paradigms for designing future cybersecurity frameworks.