超飽和設(shè)計數(shù)據(jù)新的分析方法.pdf

1、分類號密級⑧學(xué)校代碼：10055學(xué)號：030073高礅夫凄’NankaiUniversity論文題目：超飽和設(shè)計數(shù)據(jù)新的分塹方法先乞‘p／b欲碩士學(xué)位論文MASTER。SDlSSERl≯汀ION培養(yǎng)院系：一級學(xué)科：二級學(xué)科：論文作者：指導(dǎo)教師：數(shù)學(xué)學(xué)院數(shù)學(xué)概率論與數(shù)理統(tǒng)計邵敏張潤楚教授南開大學(xué)研究生院2006年5月口新月冉AbstractAbstractAsIlpersaturateddesign(ssD)isessentiallyaf

2、actorialdesignwiththeequaloccurrenceof1evelspropertyandnomllyaliaSedfactorsinwhichthenumberofmaineffectsisgreaterthanthenumberofrunsVariableselectionplaysaninlportamroleinanalyzingdatafrom恤esupersaturateddesignfkcentlyse

3、V—eralmethodswereproposedforanalyzingthetwo一1eVelSSDsLin(2002，2003)introducedapenalizedregression，SCADmethod，whichoutperfbrmsotherap—1)roachesHowever：theSCADmethodrequiresaninitialvaluefromthestepwiseregressionInthispape

4、r：weintroducetheE1asticNetnlethodanewpenalizedregressionmethodwhichcombinestheRjdgeRegressionandLassoTheElasticNetoftenoutperformsthelassointermsofpredictionaccuracyjwhileitenjoysasimilarsparsityofrepresentationandcanset

5、tlethepnproblemlikeSSDsThea110theradvantageofElasticNetistha“tdoesnotneedanyiniti“valueButasLeng，Linand、Vahba(2004)showedthat，whenthepredictionaccuracyisusedast11ecriteriontocIloosethetuningparanleter，恤eElasticNetprocedu

6、resarenotconsistentinternlsofvariableselectiollThesimulationstudyinnlypaperalsocon矗rmst11isTbovercomethisshortcoming，Weaddasupernuouscoemcientspruning1)rocedureaftertheElasticNetmodel，andcallittheAdjustedElaSticNetproced