The Application of Partial Least Square Regression in Dimensional Reduction Analysis of Spirit Samples
Statisticians are concerned with the type of multidimensional reduction techniques applied after data collection by formulating a set of new components using original variables, but chemists are more concerned with selecting a subset of the original variables so as to reduce further experiment cost The variable importance of projection in the partial least squares (PLSR) model can be used as the indicator of the importance of each variable, thus provides a criterion for variable selection. Compared with LASSO, such method can overcome the restriction of small sample size. We present an empirical study of multidimensional reduction analysis to detect the grade of Chinese Spirit samples. The variable selection procedure is especially emphasized, and data reconstruction quality is compared with that of LASSO. We find that PLSR not only provides efficient method for data viewing, but also demonstrates its effectiveness in variable selection, which can be used as the guideline for further chemical experiments.
partial least square regression dimensional reduction variable selection reconstruction quality variable importance of projection
Zuo Chen Zhong Qiding XieYihui Li Qiu
Renmin University of China, Beijing 100872, China China Research Institute of Food and Fermentation Industry, Beijing 100027, China University of Illinois at Urbana-Champaign
国际会议
The 6th International Conference on Partial Least Squares and Related Methods(第六届偏最小二乘及相关方法国际会议)
北京
英文
360-364
2009-09-04（万方平台首次上网日期，不代表论文的发表时间）