A Novel Stacked Regression Algorithm Based on Slice Transform for Small Sample Size Problem in Spectroscopic Analysis

In spectroscopic data analysis, small sample size (SSS) problem occurs. A solution is to perform variable selection, which has been proved to be critical to improve the performance of the regression model, such as partial least squares (PLS) regression. Stacked moving window partial least squares (SMWPLS) aims to combine variable sets instead of selecting a subset to improve the model robustness. In this study, we proposed a novel weighting strategy to calculate the combination weights. Slice transform (SLT) is used to map the cross-validation (CV) weights to new weights in a piecewise linear manner. The parameters of SLT are optimized with the least-square criterion. Experiments on two near-infrared (NIR) data sets demonstrated the efficiency of the proposed SLT weighting.