Real-coded GA coupled to PLS for rapid detection and quantification of tartrazine in tea using FT-IR spectroscopy.

In this study, Fourier-transform infrared (FT-IR) spectral data was combined with variable selection methods to
measure tartrazine adulteration in tea powder. Partial least square (PLS) regression and its variants such as
backward interval PLS (BiPLS), genetic algorithm PLS (GA-PLS), and competitive adaptive reweighted sampling
PLS (CARS-PLS) for variable selection were established as calibration models for the quantitative prediction of
tartrazine. A simple and efficient real-coded GA (RCGA) was also implemented as a variant of GA-PLS regression.
The performance of these models was adjudged based on root mean square errors (RMSE) for both crossvalidation (RMSECV) and prediction (RMSEP) along with their respective correlation coefficients (RC and RP).
The developed RCGA-PLS was observed to be a robust technique to achieve a model with low RMSECV and
RMSEP values of 0.8331 and 0.923, respectively. This model uses 30 selection variables (1.19% of full variable
count) to predict tartrazine in the range of 0–30 mg/g with a correlation coefficient of 0.987. This study
demonstrated that FT-IR spectroscopy, combined with the developed RCGA-PLS procedure for variable selection
could be a robust technique for the rapid detection of tartrazine in tea samples.