A method for determining a partial least squares regression latent variable number

Abstract

The invention discloses a method for determining a partial least squares regression latent variable number, which comprises the following steps: selecting a latent variable number, and selecting N sub-training sets and N sub-test sets in one-to-one correspondence with the N sub-training sets by adopting an interactive verification method on the basis of a collected sample; Establishing N sub-models by using the N sub-training sets, and predicting sub-test sets corresponding to the sub-training sets by using the sub-models corresponding to the sub-training sets; Recording a regression coefficient of each sub-model, and calculating to obtain a stability parameter corresponding to the latent variable number according to the regression coefficient of each sub-model; Selecting another latent variable number, and continuously carrying out the step of establishing a plurality of sub-models by adopting the interactive verification method based on the sample; And determining the corresponding latent variable number as the optimal latent variable number when the stability parameter is the maximum. The curve of the stability parameters changing along with the latent variable numbers has the characteristic of ascending first and descending second, so that a worker can determine the optimal latent variable number conveniently, and a model with good stability is established.

Description

technical field [0001] The invention relates to the field of data analysis and processing, in particular to a method for determining the number of partial least square regression latent variables. Background technique [0002] Data analysis refers to the process of analyzing a large amount of collected data with appropriate statistical analysis methods, extracting useful information and forming conclusions to study and summarize the data in detail. In practical terms, data analysis can help people to make judgments so that appropriate actions can be taken. [0003] In some specific fields, the chemical data to be dealt with in modern analytical chemistry is often high-dimensional data with a small number of samples but a large number of variables. Such data is known as the "large p, small n" problem, which is a very challenging problem for statistics. [0004] Partial Least Squares Regression (PLSR) is a common method that can deal with problems with more variables than sa...

AI Technical Summary

Problems solved by technology

However, when using this standard to select the number of latent variables, it is often encountered that the root mean square error of interactive verification has been decreasing (or the coefficient of determination has been approaching 1), or the partial least squares established by different latent variable numbers Situations where the predictive power of the regression model varies little (or the difference between the coefficients of determination is small), making it difficult to determine the number of latent variables

At this time, if the number of latent variables is still determined according to this principle, it will often result in the selection of too many latent variables, which will lead to overfitting of the partial least squares regression model.

[0006] Therefore, how to avoid the overfitting situation caused by selecting too many latent variables when establishing the partial least squares regression model is a technical problem that those skilled in the art need to solve

Images