Development of prediction models for pH and total acidity of plaa-som using reflectance near infrared spectroscopy

Abstract

                    For the development of pH and total acidity (TA) prediction models, plaa-som samples (N=188) were randomly drawn from production processes with a wide range of fermentation times and recipes. The samples were divided into 2 groups for calibration (N=141) and validation (N=47). The prediction models of both pH and TA values were developed using partial least squares (PLS) regression with Savitzky-Golay and N-point smoothing pretreatment methods. The best calibration models for pH and TA had R² (0.89, 0.72), SEC (0.16, 0.50), and R²max (1.0), respectively. Hence, both pH and TA calibration models gave a reasonable fit and were able to estimate pH and TA values in the validation step. The prediction capability of the pH model showed a good correlation (R²=0.89) with a standard error of prediction (SEP) and bias of 0.19 and 0.30, respectively, while the prediction capability for the TA model showed only a moderate correlation (R²=0.72), SEP (0.50), and bias (0.32), respectively. Although the TA model displayed a lower correlation value for prediction, pH and TA models were still able to give reasonable predictions for all tested plaa-som samples. Thus the two models developed in this study have revealed potential applications for large scale plaa-som manufacture since they can predict two important acid parameters that directly correlate with sourness of the product.