Feature Selection for Security evaluation against Attack in Adversarial Environment

:NotonlyPatternrecognitionbutalsomachinelearningtechniqu eshavebeenincreasedinsecurityrelatedapplicationslikeintrusi on,spam,andmalwaredetection,althoughitssecurityagainstwe llcraftedattacksthataimstoevadedetection.Spamfilteringisoneo fthewellknownapplicationexamplesconsideredinadversariale nvironment.Inthistask,thegoalisoftentodesignfeatureselectio nagainstattackswhichmayincurduringoperation.Hereinthisp aper,Iprovideamoredetailedinvestigationofthisaspect,byshed dingsome light onthe securitypropertiesoffeatureselectionagainstevasionattacks.A lsohere IuseRandomForestClassifiertofindevasionattacks.Theability ofrapidlyevolveofchangingandcomplexsituationshashelpeditt obecomeafundamentaltoolforcomputersecurity.Evasionattac ksmayassumesthattheattackercanarbitrarilychangetheevery feature,buttheyconstrainthedegreeofmanipulation,e.g.,limiti ngthenumberofmedications,orit’stotalcost.AdversarialFeatu reSelectionarchitecturemodelare givenin thispaper.