Development of a module for anomaly detection in a network graph based on neural networks

Abstract

The rapid development of artificial intelligence is fundamentally transforming our lives and making them more convenient. In particular, anomaly detection in network traffic is one of the advances that has a significant impact on society. Traditionally, anomaly detection was carried out using statistical methods; however, with the emergence of machine learning—especially deep learning—the accuracy of anomaly detection has greatly improved, and its application scope has expanded.

Anomaly detection algorithms are methods used to identify data that significantly deviates from normal data and its behavior. Such algorithms make it possible to detect anomalies effectively and accurately in large data volumes, reduce risks and costs, and increase user satisfaction.

Purpose of the study. The purpose of this research is to analyze the capabilities of neural network models for detecting anomalies in network traffic, evaluate their effectiveness compared with classical security methods, and develop practical recommendations for integrating them into information security systems.

Practical significance. The results of this study can be applied to:
- Improving intrusion detection systems (IDS/IPS) through the introduction of machine learning algorithms that enhance the accuracy and speed of threat detection.
-Reducing the workload on security specialists by automating traffic analysis and minimizing false positives.
- Increasing resistance to new types of attacks, including previously unknown threats, due to the ability of neural networks to detect anomalies without predefined rules.
- Optimizing security resources through adaptive approaches that allow systems to learn independently from new data and changing network conditions.

The integration of neural network technologies into cybersecurity systems may become a new standard of protection, providing reliable defense against dynamic and sophisticated cyber threats.