机器学习十大算法：kNN

《机器学习十大算法：kNN》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、Chapter8kNN:k-NearestNeighborsMichaelSteinbachandPang-NingTanContents8.1Introduction...........................................................1518.2DescriptionoftheAlgorithm...........................................1528.2.1High-LevelDescription.........................................1528.2.

2、2Issues..........................................................1538.2.3SoftwareImplementations......................................1558.3Examples..............................................................1558.4AdvancedTopics.......................................................1578.5Exe

3、rcises..............................................................158Acknowledgments...........................................................159References..................................................................1598.1IntroductionOneofthesimplestandrathertrivialclassifiersistheRotec

4、lassifier,whichmemorizestheentiretrainingdataandperformsclassificationonlyiftheattributesofthetestobjectexactlymatchtheattributesofoneofthetrainingobjects.Anobviousproblemwiththisapproachisthatmanytestrecordswillnotbeclassifiedbecausetheydonotexactlymatchanyofthetrainingrecords.Anotherissuearises

5、whentwoormoretrainingrecordshavethesameattributesbutdifferentclasslabels.Amoresophisticatedapproach,k-nearestneighbor(kNN)classification[10,11,21],findsagroupofkobjectsinthetrainingsetthatareclosesttothetestobject,andbasestheassignmentofalabelonthepredominanceofaparticularclassinthisneighborhood

6、.Thisaddressestheissuethat,inmanydatasets,itisunlikelythatoneobjectwillexactlymatchanother,aswellasthefactthatconflictinginformationabouttheclassofanobjectmaybeprovidedbytheobjectsclosesttoit.Thereareseveralkeyelementsofthisapproach:(i)thesetoflabeledobjectstobeusedforevaluatingatestobject’scla

7、ss,1(ii)adistanceorsimilaritymetricthatcanbeusedtocompute1Thisneednotbetheentiretrainingset.151©2009byTaylor&FrancisGroup,LLC152kNN:k-NearestNeighborstheclosenessofobjects,(iii)thevalueofk,thenumberofnearestneighbors,and(iv)themethoduse