基于后驗(yàn)貝耶斯擬似然線性模型變量選擇.pdf

1、南開大學(xué)碩士研究生畢業(yè)(學(xué)位)論文姓名：年級：專業(yè)：研究方向論文題目完成日期導(dǎo)師杜君2002級概率論與數(shù)理統(tǒng)計(jì)數(shù)理統(tǒng)計(jì)、hriableselectlonvlaQuasiPosteriorB8胂ianLikellhoodforLiⅡearModcls2005年4月lo日林路教授二00五年四月十日AbstractI矗theproduceofconstructinggeneralregressionmodel8induding1inearge

2、ner以izedlincarandnonline缸regression，al盯genumberpredictor8areusuallyintroducedattheiⅡitialstageofmodellingtoTeduccpossiblemodellingbiasesInordertoenhanccthepredictabilityandreducevariance8ofestimationandpre出ction，however，

3、oneusuallydeletesomei11signi丘cantvariablesandattempttoobt出naⅡoptim“subsetofvariablesForthisproblemselectioncriteriasucllasAIC，Cp，BICandRIChaⅣe矗xeddimensionalitypen缸tiesSuchcriteriaareshawntocorrespondtosele“ionofma面m眥pos

4、teriormodelsunderimplicithyperparameterchoicesforaparticularBayesformulationButtheexistingpenalizedIikeIihoodapproachtovariablese—lectiondependsontheHkelihoodfunctionandthenisunavajlableifthedistrmutionofdataisunknownInt

5、hjsp叩erwesuggestQua8iPosteriorBa辨siannkelihoodtoselectvariable8Theseappr。achesdonotrequireanyassumptiDnondjstributionofdatarather亡haIlthe王r丑rstmomentor矗rsttwomomentsThesemethodssclectⅧriablesandestimateparameterssiⅢultan

6、eouslyThcpr。posedestiⅢatorsperformasweUastheorackprocedu工eandareasymptoticallye伍cientP壯ticular埽forlinearre擎es8ionmodel8，thev盯iabkselectionviaposteriordensityisequivalenttothevariableselectionVi8penalizedresidu“sumofsquar