Improving the prediction accuracy of the independent component regression calibration model for the quantitative analysis of the near infrared spectroscopy

Abstract EN

Independent Component Regression (ICR) has shown promise as a rapid and non-destructive regression method in the noninvasive detection of Glucose using Near-infrared spectroscopy (NIR).

In this work, the effect of several preprocessing methods on the independent component calibration model (ICR) was investigated and analyzed including the Moving Window Average filter (MWA), Savitzky Golay filter, Band pass filter (BPF), Second derivative combined with the MWA and Savitzky Golay filter and second derivative based on Savitzky Golay filter.

The results show a clear improvement in the capability of ICR model to predict the glucose concentration from aqueous solutions.

In general, the BPF filter and second derivative based on Savitzky Golay filter gave better results than the other preprocessing methods.

The standard Error of prediction (SEP), correlation coefficients r and root mean square error of prediction (MMPE) for the BPF-ICR are 14.07, 0.9898 and 12.56, respectively.