Research On Detection Of Threat Behaviors Of Complex Network Based On Probabilistic Graphic Model

Complex network threat behaviors are made of a series of attack steps which are complex,inconspicuous and intention clearly.The sponsors are usually high-tech based people with network attack skill or high-intelligent malware software.Complex network threat behaviors are very common means of attack as many network security incidents in recent years,which have become one of the focuses of the current network and information security.Based on the in-depth research about complex network threat behaviors,we proposed a complex network threat behavior detection method based on probabilistic graphic model,and implement a prototype system of complex network threat behavior detection.The main work and innovations are as follows:Firstly,we do in-depth analysis about the features and patterns of the network behaviors for modeling based on probabilistic graphic models to solve the problem of modeling for the complex network threat behaviors.We do detailed mathematical derivation in representation,inference and learning so that there could be a better description and more generalized expression,a more efficient calculation method for the calculation of the probability distribution of the edges and a more efficiently learning progress.Secondly,we propose main technology for complex network threat behavior detection to solve the problem of complex network threat behavior detection.we use offline learning method based on network behavior patterns' extraction and online detection method based on network behavior features' detection to analyze real network traffic protocol and network behavior,extract features of network behavior,train the network behavior and get network behavior patterns.Extracted patterns of network behavior can be used as complex network threat behavior detection rules and do further testing.Thirdly,we propose system construction of the complex network threat behavior based on probabilistic graphic models to solve the testing problem.We test the offline learning method and the online detection method.The results show that the validity of offline learning method and the veracity online detection method.By compared with other machine learning algorithms,we can see the detection effect of this model is better than other machine learning algorithms.