A partition-based variable selection in partial least squares regression

Partial least squares regression is one of the most popular modelling approaches for predicting spectral data and identifying key wavelengths when combining with many variable selection methods. But some traditional variable selection approaches often overlook the local or group information between the covariates. In this paper, a partition-based variable selection in partial least squares (PARPLS) method is proposed. It first uses the k-means algorithm to part the variable space and then estimates the coefficients in each group. Finally, these coefficients are sorted to select the important variables. The results on three near-infrared (NIR) spectroscopy datasets show that the PARPLS is able to obtain better prediction performance and more effective variables than its competitors.