Only the best model
for each fruit is shown on Table 2. Initially, very different trends were observed for the evolution of the calibration error (RMSECV), according to the number of LVs, between the three species (Fig. 2). For passion and tomato fruits, evolution of RMSECV with BTK inhibitor number of LVs showed no consistent trend (Fig. 2a). The behavior of the unstressed calibration error for the passion and tomato fruits was characterized by low correlation coefficients between predicted and measured values. The best PLS model developed for the passion fruit used pre-processing multiple scatter correction (MSC) and 5 LVs which provided the lowest cross validation error of 1.62 °Brix. When the model was applied to predict the 12 internal validation samples, a low correlation (R2 = 0.63) and a high error of prediction (RMSEP% = 9.8%) were found ( Fig. 2b). For tomatoes, results were similar to the results found for passion fruits ( Fig. 2c). The lowest cross validation error (0.13 °Brix) was observed for models using 10 LVs and MSC pre-processing. When the model was used to predict the 32 internal validation samples, the prediction error was 8.85% and the correlation
coefficient was 0.52 ( Fig. 2d). However, in apricot, the relationship between RMSECV and number of LVs followed a regular profile, and a good correlation was found ( Fig. 2e). The same ratio was observed by Camps and Christen (2009). The lowest cross validation error (0.69 °Brix) was observed for models using 6 LVs Selleckchem Hydroxychloroquine and MSC pre-processing followed by smoothing. A high Protein Tyrosine Kinase inhibitor correlation coefficient (R2 = 0.93) and a low prediction error (RMSEP 3.3%) were observed, when the model was used to predict the 24 internal validation samples ( Fig. 2f). Measurement of acidity-related parameters in intact fruits is notoriously
difficult (Flores, Sánchez, Pérez-Marín, Guerrero, & Garrido-Varo, 2009). Such difficulties can be observed in Fig. 3. Similarly to what was found for the soluble solids content, when the cross validation error was plotted against the number of LVs for passion fruit and tomato (Fig. 3a and c), the correlation coefficients were below 0.49 and 0.51, respectively, indicating a poor relationship between measured and predicted values for titratable acidity. The best PLS model developed for the passion fruit used pre-processing first derivative and 5 LVs, which resulted in a cross validation error of 14.69 mmol H+·100 g FW−1. When the model was used to predict the 11 internal validation samples, a low correlation (R2 = 0.49) and a high value for the error of prediction (RMSEP% = 11.4%) were found ( Fig. 3b). For tomatoes, a minor cross validation error (0.35 mmol H+·100 g FW−1) was observed for a model using 8 LVs and MSC pre-processing. When the model was used to predict the 32 internal validation samples, a prediction error of 10.43% and a correlation coefficient of 0.