Anomaly Detection in Time Series Data for Cyber Security: Integrating Supervised, Semi-supervised and Unsupervised Methods
Howard, C.C *
Department Mathematics and Computer Science, University of Africa, Toru-Orua, Bayelsa State, Nigeria.
Omamoke, L
Department Mathematics and Computer Science, University of Africa, Toru-Orua, Bayelsa State, Nigeria.
Waidor, T. K
Department Mathematics and Computer Science, University of Africa, Toru-Orua, Bayelsa State, Nigeria.
*Author to whom correspondence should be addressed.
Abstract
This research aims to address the gap between evolving cyber threats and existing detection systems. Organizations face challenges with large data streams and sophisticated attacks. Current methods have weaknesses like low labeled datasets and high false positive rates. An integrated approach combining supervised, semi-supervised, and unsupervised detection paradigms is needed to address modern cybersecurity threats. Some of the traditional techniques were examined, weighing their pros and cons, along with real-world applications, where cybersecurity threats pose significant risks is applicable such as Banking Networks, Military Networks, Hospital Networks, Trading Platforms, Power Grids and Utilities, Large super market services, Automotive Industry, etc. The experimental findings of this study using R package reveal that using ensemble approaches—where multiple detection methods work together—can significantly outperform single-method strategies, cutting false positive rates by as much as 37% and boosting detection accuracy by 24% across a range of attack types. The paper concludes with thoughts on future research paths, especially focusing on adaptive learning systems that can keep pace with the ever-changing threat landscape.
Keywords: Low labeled datasets, cyber security, adaptive learning systems, ensemble approaches, false positive rates