PLS regression variable selection for quantification of biopolymer

Published: 17 October 2024| Version 1 | DOI: 10.17632/tjdz43w9mm.1
Contributor:
BISWANATH MAHANTY

Description

Spectral data for biopolymer solution in water and cell-free supernatant are used to build PLS models. Four variable selection algorithm, i.e., GA, CARS, ABUSE, and PSO are adopted to build PLS sub-models on reduced variable space. The LOD estimation is performed on the data sets.

Files

Steps to reproduce

There are two folders, the temp folder contains the basic pls functions, and experimental data sets (X,Y, wavelength). The CARS-GA-PSO folder contains code for univariate calibration (a0_0Pollulan_single_var_CV.m), spectral data visualization (a0_1Pollulan_spectra_plot.m), full-spectrum PLS (a0_Full_PLS.m). GA, CARS, PSO, and ABUSE algorithms for variable selection (with 100 reptation) is encoded in a1_GA_matlab.m, a2_CARS_Matlab.m, a3_PSO_Matlab.m, a4_ABUSE1.m, respectively. The data analysis after variable selection is shown in c1_GA_PSO_CARS_Assesment.m, and c1_GA_PSO_CARS_analysis.m. The models are compared using d1_Compare_models_Balandaltman.m. The figure of merit (LOD) from the models are mapped to e1_PLS_FOM_MCV.m

Institutions

Karunya University

Categories

Genetic Algorithm, Particle Swarm Optimization, Quantitative Technique, Latent Variable, Sampling, Mulitvariate Partial Least Squares Regression

Licence