资源描述：
《多元数据分析.ppt》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。
1、多元数据分析刘国庆Question?Howtodetermineafairmarketvaluefortheproperty?Inbuildingthemodel,thegoalistopredictthemarketvalueoftheLeslieSaltproperty,usingtheinformationavailable.DataLeslieSaltData.txtRegressionAnalysisItisusedtoexploretherelationshipbetweenasetofin
2、dependentvariables(X’s)andasingledependentvariable(Y).Aregressionmodelisalinearcombinationofindependentvariablesthatcorrespondsaspossibletothedependentvariable.RegressionforansweringquestionsHowcanwedescribetherelationshipbetweenthedependentvariableandth
3、eindependentvariables?Istherelationshipdescribedbythemodelstatisticallysignificant?Whichindependentvariablesaremostimportant?Howwelldoesthemodelgeneralizetoobservationsoutsidethesample?资料预处理大多数的销售价格低于$10,000,但是有一些价格甚至高于$20,000or$30,000每亩；一般地，这种价值的差别会带来问题
4、。为此，采用log函数。一些特殊的点需要排除VariablesDescriptionVariables.txtCorrelationMatrixCorrMatData.txt分析从表3中可以看出与log(Price)高度关联的变量是Elevation,Sewer,Date,Flood.从表格中我们也可以看到在独立的变量之间也存在着相关性。比如，Elevation,County.回归模型ResultsofRegressionResultsRegress.txtHowgoodistheFit?Howgood
5、istheFit(continueI)OneofthedrawbacksofR^2isthatwheneveranindependentvariableisaddedtothemodelitalwaysincreases,nomatterhowsmallthecontributioninfit.Ingeneral,whenbuildingmodels,onewantstomakeatrade-offbetweenparsimonyandimprovementinfit.HowgoodistheFit(c
6、ontinueII)IsitSignificant?Testtheerrortermsarenormallydistributed;thatisTestingthemodelF-TestFTest.txtThecriticalvaluefortheF-distributionatthe0.01levelisF(4,26)=4.14,whichsuggeststhatourmodelishighlysignificant.DetectingproblemswiththemodelOnemeasureofm
7、ulticollinearityiscalledtheconditionindex(CI)Heteroscedasticity(异方差性)Wehaveassumedthattheerrortermsallhavethesamevariance.Thisassumptionissometimescalledhomoscedasticity.Whentheassumptionisviolated(i.e.,thevariancesarenotallthesame),wehavewhatiscalledhet
8、eroscedasticity.Weightedleastsquares(WLS)InfluentialObservationsSometimesregressionmodelresultscanbeinordinatelyinfluencedbyoneorafewobservationsinthedata.Thistypeofobservationasanoutlier.DFBETAsWecanlookattheimpactoftheob
