A segmented PLS method based on genetic algorithm
Partial least square regression (PLS) establishes a multivariate linear regression model, which has low ability to make a nonlinear relationship between independent variables and dependent variables. Therefore, traditional PLS models are not able to reflect the nonlinear attributes of the sample sets very well. In order to obtain a nonlinear approximation in the multivariate analysis, a segmented PLS model based on genetic algorithm (GS-PLS) is proposed. In this method, the optimal segmentation mode of samples was directly sought based on the genetic algorithm, then, a PLS model was established for the sample subset, and a smooth continuous nonlinear PLS model was obtained with the interpolation function. The effectiveness of GS-PLS model was verified by a simulation dataset and three near infrared spectroscopy datasets of tablet, corn and meat. The results show that the proposed GS-PLS method is more robust than the segmented PLS model based on the iterative algorithm. Therefore, it has a stronger modeling ability for analyzing nonlinear data. In addition, the improvement effect of the proposed method for the PLS model was analyzed in this study. It was proven that the proposed method was a valid method to increase the effectiveness of PLS models for processing nonlinear data. The method also shows a significant improvement when the nonlinear relationship is the main factor restricting the effect of the PLS model.