A bootstrapping soft shrinkage approach for variable selection in chemical modeling

Grade Identification of Tieguanyin Tea Using Fluorescence Hyperspectra and Different Statistical Algorithms

“…Bootstrapping Soft Shrinkage (BOSS) combines the concepts of Weighted Bootstrap Sampling (WBS) and Model Population Analysis (MPA) (Deng, Yun, & Cao, ). The weight of the variable is determined by the absolute value of the regression coefficient (RC).…”

Chemometrics in instrumental analysis of complex systems—in honor and memory of Yi‐Zeng Liang

“…An application of the MPA idea is to establish variable selection methods, such as margin influence analysis for selection of discriminating genes for cancer classification . Based on MPA, Yun and Deng proposed a series of methods with the approach of the binary matrix sampling method, including iteratively retain informative variables (IRIV), variable combination population analysis (VCPA), and the variable iterative space shrinkage approach (VISSA), which optimizes variable space in a stepwise manner. Promising prediction ability was illustrated for the modeling of NIR data of soy, diesel fuel, beer, corn, and wheat samples.…”

An efficient variable selection method based on random frog for the multivariate calibration of NIR spectra

“…These procedures can be distinguished from each other either based on single wavelength selection or wavelength interval selection. Typical single wavelength selection methods include forward selection, 13 backward elimination, 14 stepwise selection, 15 uninformative variable elimination (UVE), 16 Monte Carlo-based UVE (MC-UVE), 17 successive projection algorithm (SPA), 18 iterative predictor weighting (IPW), 19 competitive adaptive reweighted sampling (CARS), 20,21 random frog (RF), 22 recursive weighted partial least squares (rPLS), 23 iteratively retaining informative variables (IRIV), 24 variable combination population analysis (VCPA), 25 variable iterative space shrinkage approach (VISSA), 26 bootstrapping so shrinkage (BOSS), 27 latent projective graph (LPG), 28 methods based on optimization algorithms, such as genetic algorithm (GA), 29,30 particle swarm optimization (PSO), 31 and ant colony optimization (ACO), 32 and methods based on regularization, such as least absolute shrinkage and selection operator (LASSO), 33 elastic net (EN) 34 and sampling error prole analysis-LASSO (SEPA-LASSO). 35 Due to the fact that the absorption band of a functional group corresponds to a relatively short wavelength band in the spectrum, it makes more sense to nd the most useful spectral band interval instead of a single spectral point and it is also easier to obtain a stable model and explain the model.…”

“…45 It is worth noting that most of methods mentioned above are based on model population analysis (MPA), 46 such as MC-UVE, 17 CARS, 20,21 RF, 22,43 IRIV, 24 VCPA, 25 VISSA 26,44 and BOSS. 27 MPA is a general framework for developing new procedures in chemometrics and bioinformatics. It mainly computes the statistical information of every variable from a large population of sub-models built with a large population of variable subsets which are generated by different sampling methods.…”