Research On Intrusion Detection In Access Network Based On SVDD And Density Peak Clustering Algorithm

With the coming of the information era,network technology becomes even more common and broadband access network enter a high-speed development stage.The application of social communication,e-commerce,e-mail and other forms of network technique brought great convenience to human lives,meanwhile,the security of access network has gradually become one of the most prominent problems,which draws worldwide attention.Intrusion Detect System(IDS)has became an indispensable technical tools to guarantee the information security because of its unique advantages.However,The data collection points of access network are too large,the type and content of services are too complex,however traditional IDS based on rule or event detection has poor accuracy,slow speed and strongly relys on rules or events.it is unable to meet existing demand.In recent years,the intelligent machine learning has been gradually improved.And the application of this technique in IDS can exactly solve the existing problem.In particular,Support Vector Data Description(SVDD)has great advantage in dealing with large-scale data,high dimension and nonlinear dataset in single classification problem.However,the research time of SVDD algorithm is short,and the theoretical research is still in its infancy.and the prediction accuracy of SVDD algorithm is obvious slanted to majority-class(the accuracy of less class is far less than that of majority-class),therfore more work need to be done.Aiming the above problems,this thesis propose an algorithm which combining Developed Peak Density Clustering algorithm with SVDD algorithm named DDPC-SVDD.This algorithm try to describe a large-scale and loose dataset by several compact sub clusters.Although,traditional DPC can obtain several convex type clusters,the value of cutoff distance(dc)is chosen experientially resulting in the instability of the clustering results.To improve the DPC algorithm,this thesis introduce an Adjusted Silhoustte coefficient(ASIL)which also shows good performance in dataset with noise.Search for the optimized value of dc within the given threshold by ASIL,then get the best clustering results.In order to generate perfect classifier,the parameters in SVDD is seachered by adaptive mutation Particle Swarm Optimization algorithm(PSO).The process of training access network intrusion dataset is executed completely automatically without set the number of cluster(k)and SVDD parameters.Simulation demonstrates that the ASIL index can evaluate the effect of clustering accurately,the accuracy of DDPC algorithm is obviously higher than other clustering algorithms,and the DDPC-SVDD algorithm is not only works well in standard UCI dataset,but also obtains satisfied results in unbalanced Kdd Cpu 1999 dataset(classic network intrusion sample dataset).