Comparison of network security situation prediction with different methods
Comparison of network security situation prediction with different methods

W. He and H. Hu, “Composite Prediction Method of Network Security Situation Based on CEEMD and Time Series Estimation”, Revista de la Facultad de Ingeniería U.C.V., Vol. 31, N°3, pp. 49-58, 2016, doi:10.21311/002.31.3.05

Abstract

“Composite prediction method of network security situation based on CEEMD and time series estimation has been proposed. Network security information has been decomposed based on the completeness of CEEMD and low frequency linear part as well as high frequency non-linear part has been attained ARIMA modeling and estimation have been adopted for linear data, adaptive radial basis function prediction has been made to non-linear part and finally reconstruct these two parts to form the final prediction model. Simulation result has shown that compared with other prediction methods, making prediction for the value of network security situation with adoption of the method in this paper can greatly improve the prediction accuracy. Make prediction error mainly concentrates on high frequency component of the first few orders through CEEMD decomposition of original situation value; the adoption of frequency division composite prediction method not only saves the overall training and prediction time, but also decreases prediction error effectively. The prediction result can follow up the changing trend of network security more accurately and provide useful reference to the management of network security.”

INTRODUCTION

“With the rapid development of internet technology in recently years, endless network security events have aroused high attention of people. Most of current network protection measures are passive defense and the rising of network security situation prediction in recent years has changed the network security management from passive to active, therefore, network security degree uses network situation value to evaluate commonly. There are certain rules in the changing of network security situation and for some time in the future, the trend is in close relationship with the happened network security situation. Predicting network security situation in the future based on existing network security situation is very popular for current research. The prediction of network security situation is to transfer existing discrete monitoring points into continuous time series and make regression analysis prediction based on the changing trend of this series (Elattar et al, 2010; Fan and Zhou, 2013; Bass, 2000; Shi and Xie, 2013; Dutt et al, 2013; Bass, 1999). The core problem of network security situation is the problem of prediction accuracy and designing a more complete prediction technology is of important practical significance for network information system to make timely and safe early warning. At present, there are many scholars making research on the prediction problem of network security situation, in which linear evaluation methods of multiple linear regression method (Zhang et al, 2011) and polynomial iterative method (Hong et al, 2010) were adopted at earliest. This kind of method is with easy calculation method and can be realized easily, but due to the non-linear, mutation and other characteristics of network information, the prediction accuracy of linear evaluation method is low. With the deepening of research, most of literatures in recent years adopt nonlinear fitting approximation methods and the representative ones are BP neural network as well as the RBF, adaptive RBF neural network and other prediction methods (Ren et al, 2006; Chen et al, 2013; Li et al, 2014; Wang et al, 2007; Ferreira et al, 2003; Tang and Yu, 2009) extended on the basis of BP neural network. However, when the target information is relatively complicate, neural network method presents shortages of difficulty in parameter selection, slow convergence rate and over-fitting etc (Deng et al, 2010). In addition, the existing literature also proposes support vector machine, gray theory and particle swarm optimization etc (Shi et al, 2011; Chen et al, 2006; Onwubiko, 2009). As network information is affected by various factors, most of the existing researches describe the partial information and become difficult to evaluate its changing trend in a comprehensive and accurate way.”


Interested researchers may download the full paper in PDF format here.