The Research Of Intrusion Detection Based On Feature Selection

With the widespread application of cumputer and network techonlogy, network security issues become more and more serious. The methods and intrusion tools of network attack, have been updated quickly, so the current network firewall and other passive security mechanisms could do nothing about it for many attacks.Intrusion detection technology as a proactive defense technology, to make up for the shortcomings of traditional security technology. Therefore, it has important significance to make the research of intrusion detection technology.The one of the major problems that the current practical intrusion detection system facing in is the detection speed is so lowly, it can not deal with massive data network in time,so caused a large number of false data positives and omissions, Under the condition of ensuring the accourcy,to develop a faster detection lightweight intrusion detection system has become a hotspot of current research. Many researchers have constitute lightweight intrusion detection system through made the efficient classifier, but the progress of this work is slow, and the effect was not obvious.Normally, if the feature vector contains the sufficient types of information that the Intrusion-detection system can classified correctly through the classifier, and whether the feature vector contains enough types information is difficult to determine,To improve the recognition rate of the classifier,the researcher always to extract the feature information maximize, the results is not only space dimension of feature increasing, but also have a larger relevance and redundancy among the features.This has led to great difficulties to realize of the classifier based on these characteristics, And therefore do need to reducing the dimension of feature space under the premise of accuracy of classification(Feature Selection), At present, for intrusion detection, the optimal feature selection algorithm is the algorithm that based on support vector machine and genetic algorithm (ie, FSSG algorithm), but the featrue selection speed of the algorithm have some probles,and the correct rate of the algorithm have a number of inadequate in intrudection detection,So,this article from another point of view, Invention an efficient feature selection algorithm to build lightweight Intrusion Detection System.In this paper, had a in-depth study of the genetic algorithm and simulated annealing algorithm, and base on the specific applications of intrusion detection systems, optimized the specific parameters of these two algorithm;analyzed their strengths and weaknesses, after studying the possibility of combining them, then formed a new search strategy for feature selection, The mixed search strategy overcomed the shortcomings of the separate two algorithms, and retained the strengths of both algorithems. Had a very good performance.On the broad framework of mixed search strategy, introduce non-linear constrained support vector machine optimization model as a evaluation criteria of feature selection, and form a new feature selection algorithm (FSGSOS), use this feature selection algorithm to build light weight model of intrusion detection.Use the standard data sets of intrusion detection----KDD1999, make large number of experiments, the Results show that the algorithm proposed in this paper have better performances.Firstly,verify the feature selection algorithm can accelerate the pace of feature selection; Secondly,verify that the intrusion detection system combined the feature selection algorithm have better performance than the the intrusion detection system not combined feature selection algorithm in the modeling time, detection time, detection rate;Lastly,experiments verify that the intrusion detection system that combined the feature selection algorithm popused in this paper have better right detect rate than the intrusion detection system combined typical feature selection algorithm (FSSG) in present.