Network Traffic Detection Based on Histogram and Self-similarity Matrix

Abstract

This paper introduces a new network traffic detection method to detect network malicious traffic in different network architectures. In most network traffic data sets, training data and testing data have the same network architecture and traffic features, like the same IP address distribution, the same IP port utilization, etc. But in real network detection, feature differences between training data sets and target network cannot be ignored. In this paper, we use histogram and self-similarity matrix to express these feature differences and keep the traffic features in use at the same time.This method could learn from anomaly network samples and detect real network traffic with feature drift, zoom, and other variants.