基于自適應(yīng)聚類算法的DDoS攻擊檢測方法研究.pdf

1、湖南大學(xué)碩士學(xué)位論文基于自適應(yīng)聚類算法的DDoS攻擊檢測方法研究姓名：李少東申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：計(jì)算機(jī)應(yīng)用技術(shù)指導(dǎo)教師：李麗娟20110522碩上學(xué)位論文AbstractWithrapiddevelopmentofthecomputernetworktechnology，computernetworkbringsUSaconveniencelife，butitalsoprovidesmoreconditionsfornetworka

2、ttacksInrecentyears，asthemostcommonofnetworkattacks，DoS(DenialofService)attackandDDoS(DistributedDenialofService)attackhaveplayedanincreasinglyimportantroleinnetworkattacks，andhasbecomethebiggestthreatentonetworkrecently

3、Inthispaper，webrieflydescribethesituationofnetworksecurity，studyandexposethetheory，classificationandresearchsituationoftheDoSandDDoSrespectively，introducetheexistingDDoSattackdetectionmethodsandtheiradvantagesanddisadvan

4、tages，predictthefuturetrendsofDDoSattackdetectionmethodspointoutthattheDDoSattackdetectionbasedondataminingisoneofthedetectiondirectionsinthefutureWedescribesomeknowledgeoftheclusteranalysis，mainlyanalysisthekmeansalgori

5、thm，somedisadvantagesofthealgorithmareasfollowing：(1)itassumesthenumberofclusterisknownbeforewebegintocluster，butinfactitisdifficulttodetermineinadvance；(2)owingtoitissensitivetotheselectionofinitialclustercenterandthedi

6、stributionofclusterdata，SO，iftheinitialclustercenterisselectedimproperly，therewillbegreatdifferencesbetweentheclusteringresultandtheactualresultInthispaper，weproposeanadaptiveclusteringalgorithm，whichcanobtaintheoptimalc

7、lusternumberbyusingdynamicindexandobtaintheinitialclustercenterbyusingrandomsubset，thenwetestitontheUCIdatasetExperimentresultsshowthatouralgorithmcanproperlygettheclusternumberandhaveabettercapabilitythanthekmeansalgori

8、thmCombinedwiththeadaptiveclusteringalgorithm，wedesignaDDoSdetectionmethodbasedontheadaptiveclusteringalgorithm，itmainlyincludethetrainingdatapretreatment，thekeyattributesextraction，thedataminingmoduleandthedecisionmodul

9、eWealsodescribethefunctionandimplementationofeachpartindetail；usetheLLSDDoS202datasettodoourexperiment，theresultsshowthatithasagooddetectionrateandalowfalserateWecomparethealgorithmwiththekmeansalgorithmintermsofdetectio