资源描述：
《基于机器学习方法的网络流量分类.研究》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。
1、ABSTRACTWiththedevelopmentandpopularizationofinternet,networktrafficdatahasbeengrowingataremarkablerate,whichbringstheconveniencetothepeople,yet,makesthemtofacehowtocarryontheanalysiseffectivelytothesemassdataandimprovetheInternetQualityofServiceaswellaspromotingInternetintoaquickerandbetterdevelo
2、pment.Thispapermainlydoesresearchintrafficclassificationbasedonmachinelearningmethodandrelatedtechniques,includingsnifferingthenetworktraffic,generatingthestatisticalfeatures,assigningtheflowexample,featureselection,andclassifyingapplicationtypeofnetworktraffic.Innetworktrafficclassificationsbased
3、onthemachinelearningmethod,gainingthenetworkflowssample,includingtrainingexampleandtestexample,isveryimportant,firstly,weobtainthenetworkpacketsbysniffering,andclassifythegatheringnetworkpacket’sintonetworkflowsaccordingtofivetuples,secondly,afterintegratingtheinformationfromPacket-LevelandFlow-Le
4、vel,andanalyzingthepacket’sattributes(size,count,time,flag)andflow’sattributes(time),37statisticalfeaturesaregeneratedandthefeaturevectorisformedwhichrepresentseachnetworkflow.Finally,weintegrateport-based,payload-based,protocolanalysistoassignnetworksample,andobtainanautosampleassignmentsystemwit
5、hhighprecision.Innetworktrafficfeatureselection,weproposeamethodoffeatureselectionbasedonfeaturedistanceandgeneticalgorithm.Thismethodcanfindthegoodinitialcommunityofgeneticalgorithmeffectively,sowecanfindagoodfeaturessetinlittleiterativenumberoftimesofthegeneticalgorithm.Theexperimentresultindica
6、testhemethodcanreducethefeaturesquantityinordertoreducethestudiesandclassificationtime,inadditiontoremovenon-correlatedortheredundantfeaturestoincreaseclassificationaccuracy.Inclassificationbasedonmachinelearningaspect,wedoresearchinclassifyingusualapplicationtypesofnetworktrafficbyusingsixmachine
7、learningmethods,andanalyzetheexperimentalresult.TheexperimentalII万方数据resultindicatesthatclassifierbasedonmachinelearningmethodcanavoiddrawbackofthetraditionalnetworktrafficclassificationapproachtoclassifythenetwo
