[Variable selection methods combined with local linear embedding theory used for optimization of near infrared spectral quantitative models]

Guang Pu Xue Yu Guang Pu Fen Xi. 2012 Dec;32(12):3208-12.
[Article in Chinese]

Abstract

Variables selection strategy combined with local linear embedding (LLE) was introduced for the analysis of complex samples by using near infrared spectroscopy (NIRS). Three methods include Monte Carlo uninformation variable elimination (MCUVE), successive projections algorithm (SPA) and MCUVE connected with SPA were used for eliminating redundancy spectral variables. Partial least squares regression (PLSR) and LLE-PLSR were used for modeling complex samples. The results shown that MCUVE can both extract effective informative variables and improve the precision of models. Compared with PLSR models, LLE-PLSR models can achieve more accurate analysis results. MCUVE combined with LLE-PLSR is an effective modeling method for NIRS quantitative analysis.

Publication types

  • English Abstract