Research On Intrusion Feature Selection Algorithms Based On ReliefF

The emergence of the Internet, not only to provide people a mutual exchange of goodlearning platform, shortens the distance between people, but also put a lot of threat back tous,hackers use various means to destroy the information system security,how to effectivelyguarantee information,availability and confidentiality, become the various challenges faced bythe industry. Intrusion detection technology as a kind of active defense, has made greatachievements in the network information security system. But as the information society hasentered the age of big data,the attribute of the information transform from the lowdimensional to higher dimensional,data showing a characteristics of large-scale, highdimension, high noise and high complexity, resulting in the "Curse of dimensionality".Putforward a severe challenges to the ability of the computing and the requirements of real-timeof the traditional intrusion detection system.As an important technology in intrusion detection, anomaly detection is well known by itcan detect unknown attack.As a typical effective anomaly detection algorithm,the SVM hasgood classification effect.But the characteristics of the data with high dimension and noisedata, seriously affected the efficiency of the SVM classification.In order to optimize thedetection model,and to reduce the complexity,our work are based on ReliefF featureselection algorithm, and introduced the basic concept of feature selection,the typical steps andthe application research status. And simple introduces the basic concept and the status offeature selection and intrusion detection, including intrusion detection model and the basicflow,and summarized and classified the intrusion detection.At the same time according to theshortcomings of intrusion detection, puts forward the future development direction ofintrusion detection.Paper studied ReliefF feature selection algorithm and its application in the field ofintrusion detection, and put forward the application and the mapping relatinship of intrusionfeature, Combined with the characteristics of network intrusion detection dataset with highsimilarity, And put forward improved Re-ReliefF algorithm in view of the traditional ReliefF applied in intrusion detection.optimization mainly aimed at the feature weight calculationmethod and combination with intrusion detection feature.In order to get better detection result, we use feature selection algorithm to process the dataof SVM classification algorithm.Experimental results show that, the improved Re-ReliefFalgorithm in all aspects of performance are improved.And after dealing with the Re-ReliefFalgorithm of data sets, the impact is not big to the classification effect of SVM, it can greatlysave the time of testing：the processed data set through the Re-ReliefF algorithm,have littleimpact on the result(accuracy) of the SVM classifier,but can save test time greatly.