📝 SummarXiv

Abstract

The approach of Near-Infrared Spectroscopy (NIRS) together with Chemometric techniques are used in order to analyze sixty (60) tablets of paracetamol of different batch numbers in the local markets in Bamako. The primary objective is to model these samples by doing multivariate regression computation. Prior to this, various statistical pretreatment methods such as second derivative (SD) correction, first derivative correction (FD), multiple scattering correction (MSC), smoothing the spectra (smooth), the standard variate normalization (SNV) correction and some combinations are performed. The Partial least square (PLS) regression on the SNV data preprocessing resulted in the detection of two outliers. Additionally, the presence of nonlinear effects is conducted. Its presence compels one to consider nonlinear regression such as the artificial neural network ANN or relevance vector machine RVM. A supporting fact of the use of these types of regressions is that ANN regression applied to the spectra under consideration significantly improves the RMSEP and the relative standard error of prediction RSEP. To further analyze the samples, the selection of wavelengths based on the p-value approach proved its usefulness in this investigation. The best calibration of the PLS multivariate regression model is obtained with the MSC combined with FD correction, and its statistical values for the fourteen wavelengths, having the smallest individual p-value, are R2= 85.26%, RMSEP= 2.38x10-4 and finally RSEP=1.45%.