Prediction of sucrose in chocolate products employing NIR spectroscopy and multivariate calibration

Considering the lower acquisition value and the practicality of obtaining data using a portable instrument (NIR-Portable), it is crucial to verify whether it offers spectra with prediction quality at least equal to those on a bench (NIR-Bench), which achieve higher spectral resolution. This study aimed to compare a NIR-Bench and a NIR-Portable to build and validate multivariate calibration models to predict sucrose in chocolate products. In Experiment 1, we evaluated cocoa mixtures with different concentrations of sucrose; in Experiment 2, fructose was added to the mixture. We built prediction models from the sample spectra via PLS regression. Considering both instruments and experiments, there were equivalent performances for the sucrose prediction models (RMSEP from 0.0597 to 0.0811 and rp2 from 0.9797 to 0.9943). A linear correlation existed between actual and predicted values ​​(0.9898 < rp < 0.997). The NIR-Portable presented, on average, greater relative errors (RE) without a predominance of greater RE per sample (p-value = 0.125). In the same spectral range, the rp was greater than 0.9874. There is no predominance of larger REs for the portable NIR (p-values ​​= 0.453 and 1 – Experiments 1 and 2, respectively). In commercial chocolate products, considerable values ​​were obtained for RMSE (0.1079 and 0.1478), r2 (0.9154 and 0.9301), and r (0.9568 and 0.9644) in Experiments 1 and 2, respectively. The fitted models allowed the detection of sucrose in both experiments. Portable NIR can replace bench-top NIR with good predictive capacity in commercial chocolate products.