A Comparison of Numerical Stability of PLS1 Algorithms

• 발행기관: 한국자료분석학회
• 저자명: Jong-Duk Kim
• 간행물 정보: 『Journal of The Korean Data Analysis Society (JKDAS)』Vol.13 No.2, 611~621쪽, 전체 11쪽
• 주제분류: 자연과학 > 통계학
• 파일형태: PDF
• 발행일자: 2011.04.30

영문 초록

Andersson (2009) evaluated nine PLS1 algorithms in terms of their numerical stability and their speed. Inspired by his work, seven PLS1 algorithms were compared in terms of their numerical stability in this paper. They were (1) NIPALS algorithm by Wold, (2) Nonorthogonalized scores algorithm by Martens, (3) SIMPLS algorithm by de Jong, (4) Improved SIMPLS algorithm by Faber and Ferre, (5) Improved Kernel Algorithm by Dayal and MacGregor, (6) Direct-scores algorithm by Andersson, and (7) A new algorithm. A measure of numerical stability was proposed, and six data sets were used to compare the numerical stability of the seven algorithms. The new algorithm developed in this paper gave the best performance for all the data sets considered.

목차

1. Introduction

2. PLS Algorithms

3. Data Sets

4. Comparison of Numerical Stability

5. Concluding Remarks

References

키워드

PLS algorithms numerical stability regression vector spectroscopic data