Chemometric calibration and partial least squares /
Chemometrics is a field of science that studies the
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
1997.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=739888091&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | Chemometrics is a field of science that studies the application of statistical methods to chemistry. In chemometrics, we usually have multiple response variables (for example, spectra), and multiple explanatory variables (for example, chemical compositions). Typically, we have fewer observations than responses. Usually, we wish to predict the true explanatory variables from the responses, for example, to predict the true compositions from the spectra. Among the available methods, partial least squares regression (PLS) is frequently used. But, in cases where most of the responses are pure error, or the background region is large, PLS does not give satisfactory prediction. This occurs because the PLS method gives excessive weight to pure noise. In this dissertation, we show that in the small sample size case under reasonable conditions, PLS with variable selection works better than the original PLS in a predictive capability, develop two new variable selection algorithms, generalize the results to general prediction function and apply the results to real examples and simulations. |
|---|---|
| Item Description: | Vita. "Major Subject: Statistics". |
| Physical Description: | xiv, 146 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilms Inc. |
| Bibliography: | Includes bibliographical references: pages 142-145. |