EARTH MOVER DISTANCE SYSTEM TO DISCLOSE DOS ATTACKS

Abstract

In this internet era, security is a major issue. Out of the many attacks DoS attack is one among them. Researchers have been working on it from 1990’s. Many systems were introduced to detect DoS attacks which used machine learning approaches and statistical analysis where in, in the proposed system we use anomaly based technique. The methods used for detection are Principal Component Analysis, Multivariate Correlation Analysis using Triangular Area Map and Earth Mover Distance. For reducing the features we make use of the Principal Component Analysis technique which is the dimensionality reduction algorithm. For effective detection , finding the correlation between the obtained features is important and hence we use Multivariate Correlation Analysis along with the Triangular Area Map. Out of the many known dissimilarity measures in MinKowski-form distance LP and X 2 statistics where it evaluates dissimilarity between distributions, we make use of different approache called Earth Mover Distanc(EMD) to attain the goal which gives higher accuracy. The uniqueness of EMD makes the proposed system more capable. Evaluations are conducted using KDD cup dataset.