ReliefF functional selection for traffic anomalies detection

Abstract

This paper addresses the problem of network traffic anomaly detection using machine learning techniques and feature selection optimization. As modern computer networks become increasingly complex and data volumes grow, efficient threat detection requires fast and accurate traffic analysis algorithms. A major challenge lies in distinguishing normal network activity from potentially malicious attacks, which may be disguised as legitimate traffic.

The study explores the use of the ReliefF method for selecting key network traffic features and its impact on the performance of the Random Forest classification model. The focus is on optimizing feature selection to reduce computational costs and speed up training while preserving essential indicators of anomalous activity. Special attention is given to evaluating system performance under different feature selection scenarios, allowing for an optimal trade-off between accuracy and processing speed.

The proposed approach is designed for automated intrusion detection systems (IDS) operating in real-time, requiring rapid responses to potential threats such as DDoS attacks, port scanning, unauthorized access, and exploits. The research aims to enhance the efficiency of existing network traffic analysis methods and adapt feature selection algorithms for high-load network environments.