Therefore, variable selection was integral in this study, because it enabled the use of models that required a small number of spectral variables. The correlation coefficients (r2) for the prediction set ranged from 0.75 to 0.90 for all models, with the exception of PLS (4) second derivative (3 pts.), and iPLS (5) ( Fig. 2). Observations in the NIR spectral region for models with derivative data showed higher RMSEP values than models with raw or smoothed data, due to loss of some important spectral information when the derivative spectra were employed. Four or five latent variables were used for NIR spectra with PLS, iPLS, SPA, and GA models. The strategy for using GA models was the advantage of

employing fewer variables (774) to build http://www.selleckchem.com/CDK.html PLS models. Two outliers were excluded from the calibration set, and the best PLS model for TAC was developed by applying a smoothing with five points. For this model, the lowest root mean square error of cross validation (RMSECV) and RMSEP were 13.8 and 4.8 g kg−1, respectively. The correlation coefficient (r2) for the validation set was 0.90, and was obtained using four latent variables. Fig. 2 depicts the correlation between measured and predicted values for TAC in açaí and palmitero-juçara. The diagonal line represents ideal results; the

closer the points plot to the www.selleckchem.com/products/wnt-c59-c59.html diagonal, the better the fit to the model. Blue open circles represent calibration spectra, and solid triangles represent validation spectra. An elliptic joint confidence region (EJCR) was constructed for the slope and intercept when plotting the predicted versus actual parameter values (at a 95% confidence interval) (Fig. 3). EJCR calculations are a convenient means to ascertain

if bias exists in determination of both parameters when using the PLS (4) smoothing (5 pts.) model. The ellipse contained the expected theoretical value of 1.0 when built for TAC (Fig. 3). The figures of merit results are provided in Table 2. Accuracy values represented by RMSEC (root mean square error of calibration) and RMSEP indicated the estimated multivariate model values exhibited acceptable agreement with the reference method. Precision, at level of repeatability, was assessed by analysing five samples/ten replicates per sample, with measurements recorded on the same Tideglusib day. Acceptable results were observed for sensitivity to the parameter evaluated, considering the analytical range of each model. A direct relationship with the prediction errors was not detected for the value of the signal-to-noise ratio, which was apparently low. This result suggested the estimated LD and LQ values might be optimistic (Table 2). A rapid and non-destructive method to determine total anthocyanin content in intact açaí and palmitero-juçara fruits using NIR spectroscopy and multivariate calibration was achieved in this study.