Multi-instrument spectroscopic study for authentication of curcumin content in commercial turmeric powders using machine learning algorithms.

Adulteration during processing of turmeric powder not only causes health risks for the consumers but also affects
its quality. There is a need for rapid and non-invasive analysis of its active ingredient, curcumin, during the
supply-chain. In the present study a total six IR instruments ranging from hand-held (NIR), portable (NIR) and
standalone (FTNIR and FTIR) were used to obtain spectral data of 160 different turmeric samples. The curcumin
content quantified using HPLC procedure was used as the response variable for analytical model using machine
learning tools. Real coded genetic algorithm (RCGA) as the variable selection procedure provided most critical
variables in the sets of 10, 20, 30 and 40 variables. Sensitivity analysis has revealed the most critical fingerprint
(s) in authenticating curcumin across all the instruments. The hand-held (NIR) device with only 20 spectral
variables resulted in 93 % accuracy using SVM classifier, and RP (regression co-efficient of prediction) values of
0.970 and 0.997 using RF and XGBoost, respectively. In case of FTNIR and FTIR instruments 100 % classification
accuracy was achieved using SVM, whereas RF and XGBoost resulted in RP values greater than 0.93. This study
enables classification and quantification of curcumin content in commercial turmeric powders using nondestructive
methodology.