Determination of glucose concentration from near-infrared spectra using principle component regression coupled with digital bandpass filter

In this paper, we have investigated the use of principal component regression (PCR) combined with time domain filtering to predict the glucose concentration from NIR spectra of mixtures composed from glucose, urea and triacetin. The whole experiments were carried out in a non-controlled environment or sample conditions to show that the PCR coupled with digital bandpass filter can suppress effectively most of the experimental variation. The filters were implemented in the time domain as Chebyshev filter for different orders (1st, 2nd and 3rd) and in the frequency domain as a Gaussian bandpass filter. The response surface method was used to optimize the filter parameters and the number of factors. The use of PCR algorithm coupled with the digital filters has decreased the standard error of prediction (SEP) from 40 mg/dL for unfiltered spectra to 19.1 mg/dL for Gaussian filtering method and 15.63 mg/dL for a well-designed Chebyshev filter.