Flexible Stacked Partial Least Squares for Mid- Infrared Spectroscopy Glucose Detection

In this work, we propose a data fusion regression approach for quantitative analysis of glucose using mid- infrared (IR) spectra. First, the approach computes the variable score index. Several submodels are then generated in terms of the index from the calibration set. Finally, the ensembled regression model is created by combining these submodels. In addition, five different regression approaches from the literature are comparatively assessed. Our research shows that one model proposed achieves good performance (with a correlation coefficient of 0.94). Our conclusion is that the data fusion model can provide an accurate and robust prediction result for IR glucose measurements.