Near infrared spectrum variable selection method based on self-weighted variable combination cluster analysis

Abstract

The invention relates to a near infrared spectrum variable selection method based on self-weighted variable combination cluster analysis, which belongs to the technical field of nondestructive analysis in the field of analytic chemistry. The specific implementation process is as follows: firstly, a variable space is randomly sampled through a binary matrix sampling method (BMS); secondly, two types of information vectors (IVs) of variables, i.e., occurrence frequencies (Fre) and partial least square regression coefficients (Reg), are weighted, so that a contribution value of each spectral variable is obtained, and thereby the influence of the two types of IVs, i.e., Fre and Reg, on variable importance is taken into consideration. Finally, the variables with relatively small contribution values are deleted through an exponential decay function (EDF), and thereby feature variable selection is realized. Compared with the prior art, the method has the advantages of quickness and repeatability, and increases the prediction precision and stability of a model.

Description

technical field [0001] The invention of the method belongs to the technical field of non-destructive analysis in the field of analytical chemistry, and specifically relates to a near-infrared spectrum variable selection method based on self-weighting variable combination cluster analysis. Background technique [0002] With the development of near-infrared spectroscopy and chemometrics, variable selection technology has become a key link in the field of near-infrared spectroscopy to analyze high-dimensional data. Variable selection for spectral variables can improve the predictive ability of prediction models. Reduce spectral data dimensionality and enhance predictive model interpretability. At the same time, variable selection is also a very challenging problem, and it is a very difficult problem to find a set of optimal variable combinations as the variable space increases. [0003] Common variable selection methods at home and abroad include Uninformative variables elimin...

AI Technical Summary

Problems solved by technology

[0004] Although a large number of variable selection methods have been proposed, each variable selection method only uses one of these information vectors as the basis for judging the importance of variables, and then ignores the influence of other information vectors on the prediction model, so it is easy to generate predictions. The over-fitting phenomenon of the model, in addition, the prediction accuracy of the existing algorithm model is low, and there are many unstable factors, which will cause the instability of the prediction accuracy of the model

Images