Loading...
|
Please use this identifier to cite or link to this item:
https://nccur.lib.nccu.edu.tw/handle/140.119/100453
|
| Title: | 無母數主成份迴歸與向前迴歸的比較
Nonparametric Principal Components Regression Compared with Forward Regression |
| Authors: | 陳弘叡
Chen, Hong Rui |
| Contributors: | 黃子銘
Huang, Tzee Ming
陳弘叡
Chen, Hong Rui |
| Keywords: | 變數選取
主成份分析法
向前選取法
無母數加成迴歸模型
backfitting algorithm
B-Spline
積分絕對誤差
Variable Selection
Principal Component Analysis
Forward Selection
Nonparametric Additive Model
Backfitting Alogorithm
B-Spline
Integral Absolute Error |
| Date: | 2016 |
| Issue Date: | 2016-08-22 10:42:31 (UTC+8) |
| Abstract: | 在一般線性迴歸模型當中，當樣本數大於變數個數時候，我們是以最小平方法來估計迴歸參數。然而當變數個數大於樣本個數時，會造成共線性問題，導致參數無法估計，無法確認個別自變數對依變數有多大影響。為了解決共線性問題，我們透過變數選取來選取重要的變數，選取方法包含主成份分析法 (PCA)及向前選取法 (FS).
我們使用的模型為無母數加成迴歸模型，透過 backfitting algorithm 來估計整個迴歸函數，個別函數則以無母數方法，使用B-Spline 來估計。我們把兩種選取方法應用在無母數加成模型裡，以積分絕對誤差為判斷標準，透過不同變數及不同生成模型類型的模擬，來判斷哪種配適選取方法較合適。模擬結果可以發現，多數情況下，FS的表現比PCA來得好。
In a general linear regression model, when the sample size $n$ is greater than the number of variables $p$, it is common to use the least squares method to estimate the parameters in the regression model. When $n<p$, the parameters in the regression model cannot be estimated due to collinearity, so it is necessary to perform variable selection before estimating the parameters. In this thesis, I compare two variable selection methods in nonparametric additive regression. The first method is based on principal component analysis (PCA), and the second method is based on forward selection (FS). The integrated absolute error is used to evaluate the performance of these two methods in simulation studies. The simulation results show that FS performs better than PCA in most cases. |
| Reference: | [1]陳順宇.多變量分析,四版,華泰書局.收稿日期:民國99年,10,2005.
[2]ZarahKatharinaSaxilAndersen.Boligpriseffektenafdenkben-havnskemetro.2009.
[3]ErnielBarriosandJenniferUmali.Nonparametricprincipalcomponentsregression.InProceedingsofthe58thWorldCongressoftheInternationalStatisticalInstitute.
[4]AndreasBuja,TrevorHastie,andRobertTibshirani.Linearsmoothersandadditivemodels.TheAnnalsofStatistics,pages453–510,1989.
[5]NormanCliff.Theeigenvalues-greater-than-oneruleandthereliabilityofcomponents.Psychologicalbulletin,103(2):276,1988.
[6]CarlDeBoor.Oncalculatingwithb-splines.JournalofAp-proximationtheory,6(1):50–62,1972.
[7]MAEfroymson.Multipleregressionanalysis.Mathematicalmethodsfordigitalcomputers,1:191–203,1960.
[8]JianqingFanandJianchengJiang.Generalizedlikelihoodratiotestsforadditivemodels.2005.
[9]JeromeHFriedmanandWernerStuetzle.Projectionpursuitregression.JournaloftheAmericanstatisticalAssociation,76(376):817–823,1981.
[10]HaroldHotelling.Analysisofacomplexofstatisticalvariablesintoprincipalcomponents.Journalofeducationalpsychology,24(6):417,1933.
[11]KarlPearson.Liii.onlinesandplanesofclosestfittosys-temsofpointsinspace.TheLondon,Edinburgh,andDublinPhilosophicalMagazineandJournalofScience,2(11):559–572,1901.
[12]IsaacJacobSchoenberg.Contributionstotheproblemofap-proximationofequidistantdatabyanalyticfunctions:Partb—ontheproblemofosculatoryinterpolation.asecondclassofanalyticapproximationformulae.QuarterlyofAppliedMathe-matics,4(2):112–141,1946. |
| Description: | 碩士
國立政治大學
統計學系
103354028 |
| Source URI: | http://thesis.lib.nccu.edu.tw/record/#G0103354028 |
| Data Type: | thesis |
| Appears in Collections: | [統計學系] 學位論文
|
Files in This Item:
| File |
Description |
Size | Format | |
|---|
| 402801.pdf | | 940Kb | Adobe PDF2 | 240 | View/Open |
|
All items in 政大典藏 are protected by copyright, with all rights reserved.
|
