Please use this identifier to cite or link to this item: http://scholarbank.nus.edu.sg/handle/10635/136276
Title: CROSS VALIDATION METHOD ON WEIGHTED ISOTONIC REGRESSION FOR NONPARAMETRIC REGRESSION FITTING
Authors: ZHANG YIWEN
Keywords: cross validation, isotonic regression, local polynomial regression, bandwidth selection, PAVA, nonparametric regression
Issue Date: 9-Jan-2017
Citation: ZHANG YIWEN (2017-01-09). CROSS VALIDATION METHOD ON WEIGHTED ISOTONIC REGRESSION FOR NONPARAMETRIC REGRESSION FITTING. ScholarBank@NUS Repository.
Abstract: This thesis discusses the local polynomial regression and the isotonic regression method to solve the nonparametric regression problem subject to the non-decreasing condition. To solve the continuous isotonic regression problem, Pool Adjacent Violators Algorithm (PAVA) is used by updating the local polynomial regression estimates of a large amount of quantiles of X. Considering the importance of the bandwidth selection for nonparametric fitting, a new Cross Validation (CV) method that incorporates the non-decreasing condition is proposed. Furthermore, the simulations are conducted to evaluate the performance of this CV method in comparison with the plug-in bandwidth selection method in the regression estimations. By obtaining the outcomes of L1 distance, its percentage improvement, and confidence interval bands, we conclude that the proposed CV bandwidth selection method improves the performance if sufficient data is provided. The proposed method in the weighted isotonic regression estimation outperforms others in this nonparametric regression problem because they both contribute to factor in the monotone constraint.
URI: http://scholarbank.nus.edu.sg/handle/10635/136276
Appears in Collections:Master's Theses (Open)

Show full item record
Files in This Item:
File Description SizeFormatAccess SettingsVersion 
ZhangYW.pdf411.95 kBAdobe PDF

OPEN

NoneView/Download

Page view(s)

108
checked on Dec 6, 2018

Download(s)

494
checked on Dec 6, 2018

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.