On the number of partial least squares components in dimension reduction for tumor classification

Authors:
Xue-Qiang Zeng;Guo-Zheng Li;Geng-Feng Wu;Hua-Xing Zou
Affiliations:
Computer Center, Nanchang University, Nanchang, China and School of Computer Engineering & Science, Shanghai University, Shanghai, China;School of Computer Engineering & Science, Shanghai University, Shanghai, China;School of Computer Engineering & Science, Shanghai University, Shanghai, China;Computer Center, Nanchang University, Nanchang, China
Venue:
PAKDD'07 Proceedings of the 2007 international conference on Emerging technologies in knowledge discovery and data mining
Year:
2007

Citing 2
Cited 1

Text classification based on partial least square analysis

Proceedings of the 2007 ACM symposium on Applied computing
Orthogonal projection weights in dimension reduction based on Partial Least Squares

International Journal of Computational Intelligence in Bioinformatics and Systems Biology

Irrelevant gene elimination for Partial Least Squares based Dimension Reduction by using feature probes

International Journal of Data Mining and Bioinformatics

Quantified Score

Hi-index	0.00

Visualization

Abstract

Dimension reduction is important during the analysis of gene expression microarray data, because the high dimensionality of data sets hurts the generalization performance of classifiers. Partial Least Squares (PLS) based dimension reduction is a frequently used method, since it is specialized in handling high dimensional data set and leads to satisfying classification performance. This paper investigates the influence on generalization performance caused by the variation of the number of PLS components and the relationship between classification performance and regression quality of PLS on the training set. Experimental results show that the number of PLS components for classifiers can be automatically determined by regression quality of PLS latent variables.