Multivariate correlation analysis technique based on Euclidean distance map for network traffic characterization

The quality of feature has significant impact on the performance of detection techniques used for Denial-of-Service (DoS) attack. The features that fail to provide accurate characterization for network traffic records make the techniques suffer from low accuracy in detection. Although researches have been conducted and attempted to overcome this problem, there are some constraints in these works. In this paper, we propose a technique based on Euclidean Distance Map (EDM) for optimal feature extraction. The proposed technique runs analysis on original feature space (first-order statistics) and extracts the multivariate correlations between the first-order statistics. The extracted multivariate correlations, namely second-order statistics, preserve significant discriminative information for accurate characterizations of network traffic records, and these multivariate correlations can be the high-quality potential features for DoS attack detection. The effectiveness of the proposed technique is evaluated using KDD CUP 99 dataset and experimental analysis shows encouraging results.