Optimal variable selection for Fourier transform infrared spectroscopic analysis of starch-adulterated garlic powder

• Fourier transform infrared spectroscopy was used for prediction of cornstarch adulterated garlic powder.
• Partial least square regression was used to develop model.
• Optimal variables were selected using model-based variable selection method.
• Applicability of the developed analytical method for analysing powder sample was demonstrated.

Fourier transform infrared (FT-IR) spectroscopy in combination with partial least-squares regression (PLSR) and variable selection methods was used for the prediction of cornstarch adulteration in garlic powder. PLSR was applied to all collected spectral data (n = 1742) for pure and adulterated samples (1–35 wt% starch). Model-based variable selection methods combined with variable importance in projection (VIP) values and selectivity ratios (SR) allowed selection of 348 and 34 optimal variables from the full set of 1742 variables, respectively. The PLSR model was also constructed using individual VIP or SR selected variables. The PLSR predictive value for the VIP-based model exhibited a correlation coefficient value Rp2 of 0.95 with a standard error of prediction (SEP) of 2.56, better than the models developed with all variables (Rp2=0.93   and   SEP=2.66) or SR-selected variables (Rp2=0.89   and   SEP=3.32). The selected variables from both the VIP and SR methods reveal the starch-related absorptions. The convincing results demonstrate the potential of model-based variable selection methods to select impressive FT-IR variables. It was found that the suggested analytical method was competent to rapidly detect starch added in garlic powder, and such an approach can be adopted for the analysis of a range of adulterated samples.