Calculation of Network Security Index Based on Convolution Neural Networks

Abstract

The Network Security Index System is an important means for network security situation assessment (NSSA). Through index selection, system construction and numerical calculation, it helps network managers obtain macro perspectives on networks. However, the traditional situation assessment methods based on an index system have always had some defects, such as excessive reliance on manual intervention, high deviation, and limited scope of application. Based on summing up the existing research results, this paper combines the advantages of adaptability, effective feature extraction and complexity reduction of convolutional neural networks (CNN). By constructing the convolutional kernels which are suitable for the characteristics of an index system we extract the potential correlation features; by using a pooling technique we shrink the model scale quickly and highlight the main features; by utilizing the deep network structure of multiple hidden layers, we implement a method of calculating network security indexes based on CNNs. Finally, the feasibility and effectiveness of this method are verified by experiments and comparisons.