Application of Improved Partial Least Squares Regression Method in Microbial Culture Medium Optimization

Abstract

The invention discloses the application of an improved partial least squares regression method in the optimization of microbial culture medium, and integrates the technology of cross-validation and variable screening to reduce the potential over-fitting problem of PLSR, that is, first extract the Appropriate number of latent variables, and then use the variable screening method based on the variable projection importance technology combined with the robust criterion of the coefficient of determination of the regression model to establish a model that is simple, robust, and especially high in prediction accuracy. The VIP regression model proposed by the present invention establishes the validity of the simple, robust and high-prediction model. Can be promoted further.

Description

technical field [0001] The invention relates to the application of a multivariate statistical method, in particular to an application method of an improved partial least square regression method in the optimization of microbial culture medium. Background technique [0002] Partial least squares regression (PLSR) is a new type of multivariate statistical method because it combines the advantages of multiple linear regression, principal component analysis and canonical correlation analysis. Studies have shown that PLSR is good at dealing with multicollinearity that cannot be solved by conventional linear regression. , small samples, etc., and can use limited data to perform regression operations of multiple independent variables on multiple dependent variables at the same time. [0003] On the other hand, however, PLSR often exhibits a tendency to overfit due to the large number of latent variables that extract confounding noise. Overfitting is a statistical phenomenon, which...

AI Technical Summary

Problems solved by technology

[0003] However, on the other hand, PLSR often shows a tendency of overfitting, which is caused by the large number of latent variables for extracting confounding noise.

Overfitting is a statistical phenomenon, which generally means that the modeling data can be perfectly fitted, but the newly added data cannot be reliably predicted, that is, the fitting accuracy is high but the prediction accuracy is poor

Images