One-sided cross-validation for a model motivated by variable star data /
Nonparametric regression techniques are often sensitive to the presence of correlation between observations. A new smoothing parameter selection method, called leave-k-out one-sided cross-validation, is proposed for a model motivated by variable star data, in which observations are negatively correl...
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
2003.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=765154791&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | Nonparametric regression techniques are often sensitive to the presence of correlation between observations. A new smoothing parameter selection method, called leave-k-out one-sided cross-validation, is proposed for a model motivated by variable star data, in which observations are negatively correlated. This new method has the objectivity of leave-(2l+1)-out cross-validation (Chu and Marron 1991) and desirable features of one-sided cross-validation (Hart and Yi 1998). In this research we study the asymptotic properties and small sample behavior of the leave-k-out one-sided cross-validation bandwidth selector. A bootstrap-based test for a trend in the periods of variable stars is also developed. |
|---|---|
| Item Description: | Vita. "Major Subject: Statistics". |
| Physical Description: | xi, 96 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilm Inc. |
| Bibliography: | Includes bibliographical references (leaves 92-95). |