Abstract:

Due to widely network applications, the role of network security is becoming more and more important in computer networks. The analysis and discrimination of network data stream and intrusion behaviors is an important direction of network security research. When anomelous behavior coming from outside is detected in network, intrusion data can be treat as nonlinear disturbance which is overlay normal network data flow. Strength of disturbance is influenced by the stream of intrusion data. Therefore, we can use non-linear theory and model to construct non-linear pattern for the network data stream. Then abnormal behavior could be discovered based on parameter fitting method. In response to network intrusion detection, this paper introduces a nonlinear regression method - partial least squares to predict the network behaviors. At the same time, in the calculation of partial least squares residuals, the paper adopts the Kullback Leibler-divergence as an iterative calculation standard so as to improve the detection speed and accuracy.

Key words: divergence, partial least squares, intrusion detection, network security