Assistant detection of skewed data streams classification in cloud security

Abstract

Data stream in the cloud is characterized by imbalanced distribution and concept drift. To solve the problem of classification of skewed and concept drift data stream in cloud security, we present an one-class classifier dynamic ensemble method which aims at separating virus data, reducing the amount of data analyzed in clouds, improving the efficiency of intrusion detection in cloud security and assisting detection of virus. The proposed method is based on using K-means algorithm to adjust data distribution, makes use of interval estimation combined with AUC value to check concept drift and updates classifiers and dynamically allocates weights. Experimental results illustrate that the proposed method can achieve good classification performance on synthetic dataset and effectively separate most of the virus samples on KDDCUP'99 dataset.