Analysis of modern anomaly detection and correction methods in communication networks under high-load conditions

DOI: 10.31673/2412-9070.2025.046209

Abstract

This paper presents a comprehensive analysis of modern methods for anomaly detection and mitigation in communication networks, taking into account current challenges related to dynamic traffic scaling, high data transmission intensity, and the need to ensure stable network infrastructure performance in real-time conditions. The increasing intensity of data transmission in communication systems driven by the widespread use of streaming video, cloud computing, and mobile services leads to a higher frequency of anomalous events. This events can result in packet loss, increased latency, or complete service shutdown. This highlights the need for highly efficient detection systems capable of adapting to changes in the network environment. Traditional approaches often rely on fixed thresholds, signatures, or heuristic rules, which significantly limit their effectiveness under dynamic conditions. These systems cannot respond promptly to unknown threats or anomalous traffic patterns that were not anticipated during system configuration.
The study examines a broad range of contemporary methods, from classical statistical models to machine learning algorithms. Particular emphasis is placed on clustering techniques (e.g., k-means), Bayesian networks, tree-based models (Decision Tree, Random Forest), and time series approaches such as ARIMA and LSTM. The evaluation of these methods is based on their compliance with modern requirements, including scalability, real-time processing, integration with automated control systems, and minimization of false positives.
The analysis revealed that most existing solutions show limited generalization capabilities, high sensitivity to parameter tuning, and difficulties in adapting to variable load conditions. The novelty of this research lies in the comprehensive assessment for adopting hybrid systems that combine several diverse methods, including Autoencoders, GANs, VAEs, and Isolation Forest algorithms. This approach significantly enhances the adaptability and accuracy of anomaly detection even in complex and unpredictable environments. The results of this research are practically relevant for the further development of intelligent traffic monitoring systems that account for the described issues under critical load conditions.

Keywords: communication networks; anomalies; adaptivity; forecasting; scalability; network load; patters; machine learning; traffic analysis; hybrid models.