Intrusion Detection in NEAR System by Anti-denoising Traffic Data Series using Discrete Wavelet Transform

Abstract

The paper presents two methods for detecting anomalies in data series derived from network traffic. Intrusion detection systems based on network traffic analysis are able to respond to incidents never seen before by detecting anomalies in data series extracted from the traffic. Some anomalies manifest themselves as pulses of various sizes and shapes, superimposed on series corresponding to normal traffic. In order to detect those impulses we propose two methods based on discrete wavelet transformation. Their effectiveness expressed in relative thresholds on pulse amplitude for no false negatives and no false positives is then evaluated against pulse duration and Hurst characteristic of original series. Different base functions are also evaluated for efficiency in the context of the proposed methods.

Keywords

• discrete wavelet transform
• intrusion detection
• self-similarity
• signal denoising
• time-frequency analysis