Skip to main content
AkademIndex

Products

For developers

AkademBasesoonOpen API for the ecosystem
Latin
Article

Algorithm for Detection and Classification of Anomalies in the Traffic of Electronic Network Resources

Muhammadjon MusaevFaculty of Computer Engineering, Tashkent University of Information Technologies Named After Muhammad al-Khwarizmi, Tashkent, UzbekistanRakhmatov FurkatFaculty of Computer Engineering, Tashkent University of Information Technologies Named After Muhammad al-Khwarizmi, Tashkent, UzbekistanKholmuminov OybekDepartment of Methodology of Exact and Natural Sciences, Tashkent Region Pedagogical Skills Center, Ghazalkent City, UzbekistanKarimov NorbekDepartment of Methodology of Exact and Natural Sciences, Tashkent Region Pedagogical Skills Center, Ghazalkent City, UzbekistanAbdirazakov FakhriddinFaculty of Computer Engineering, Tashkent University of Information Technologies Named After Muhammad al-Khwarizmi, Tashkent, Uzbekistan
ABI

Abstract

This study addresses the challenge of detecting and classifying advanced cyber threats, including distributed denial-of-service (DDoS) and Stealth Attacks, in complex network environments. We propose a hybrid approach that combines machine learning models with a rule-based classifier to improve anomaly detection accuracy. The method uses Random Forest, Gradient Boosting, and Linear Regression to predict normal traffic volume with high precision (R² = 0.99, MSE = 0.000085). Key features include request rate, traffic volume, source IP entropy, flow duration, and protocol diversity. A deviation threshold of 0.5 standard deviations from predicted values effectively identifies anomalies under dynamic conditions. For classification, we introduce a rule-based system that utilizes thresholds, such as a request rate exceeding 100 requests per second for DDoS attacks or a source IP entropy of approximately 0.6907 for Stealth Attacks. This classifier identifies six types of anomalies: DDoS, Slow, Volumetric, Service Outage, Application Layer, and Stealth Attacks. The experimental results demonstrate the effectiveness of our hybrid approach. Compared to existing methods, it achieves higher F1-scores (0.97-1.00) across most attack types. Additionally, it correctly classifies 91% of real traffic (1,246,311 packets) as usual in a synthetic dataset containing 9,660 flows. The proposed method demonstrates strong performance in terms of precision, recall, and computational efficiency, making it suitable for real-time network monitoring.

Topics

Identifiers

Citations and references

Cited by 012 references
Metrics — AkademScholar · Coming soon