Improving the prediction performance of PLSR using RReliefF and FSD for the quantitative analysis of glucose in Near Infrared spectra.

This paper proposes a novel pre-processing method, Fourier Self Deconvoluted RReliefF (FSDR) that is based on combining Fourier Self Deconvolution (FSD) with the Regressional Relief-F (RReliefF) processing to improve the prediction performance of the Partial Least Squares Regression (PLSR) model in Near Infrared (NIR) spectroscopy. The FSD is used to eliminate both the baseline variations and high frequency noise from the raw spectra and the RReliefF is applied as a feature weighting algorithm. The proposed FSDR-PLSR technique is validated for the determination of glucose from NIR spectra of a mixture composed of triacetin, urea and glucose in a phosphate buffer solution where the individual component concentrations are selected to be within their physiological range in blood. The results obtained confirm that the proposed pre-processing technique improved the prediction performance of the PLSR model.