Moving Window Sparse Partial Least Squares Method And Its Application In Spectral Data

To deal with multicollinearity,partial least squares(PLS)regression is a commonly used method.Multicollinearity refers to the high correlation between independent variables,which will cause instability to the results of regression estimation.There are many methods that can be used to solve the problem of multicollinearity.Among them,partial least squares regression and principal component analysis(PCA)are relatively similar.Both methods involve extracting principal components from variables.The difference is that principal component analysis only extracted from the explanatory variable,and the partial least square method also extracts the principal component from the dependent variable,and considers the correlation between the explanatory variable and the dependent variable.This paper proposes a moving window sparse partial least squares method(MWSPLS),and compares this method with common variable selection methods through simulation experiments.The method combines the sliding interval method and the sparse partial least squares(SPLS)method.While using the sliding interval PLS regression to screen out the optimal interval,it also uses SPLS for variable selection.Compared with the traditional PLS method,the innovations of the method in this paper are:First,it can control the size of the variable interval,the sparsity and the number of components,and it has a high degree of flexibility.Second,when the variable interval contains both impurity variables and very useful variables for prediction,the combination of interval analysis and variable selection can provide a collaborative strategy for PLS modeling and interpretation.Finally,this article applies this method to seawater spectral data to establish a model.Compared with chemical measurement methods,it has the advantages of simplicity and fast measurement speed.