normal distribution, with a confidence level of 95%, assuming a
threshold value of ±2.5.
Afterwards, the models were tested to predict SSC and TA with
validation set. The best calibration models were selected based on
the highest correlation coefficient of validation (R2
) along with the
lowest RMSECV and the lowest root mean square error of prediction
(RMSEP). RMSEP was then expressed as RMSEP% corresponding
to the percentage of error of prediction calculated with
RMSEP divided by the mean values of measured quality parameters
in fruits from the validation set (Duarte, Barros, Delgadillo, Almeida,
& Gil, 2002).
RMSEP ¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn
i¼1 yi y^ ð Þi
2
n
s
ð1Þ
where: yi = known value; y^i = calculated or predicted value and
n = number of samples in the validation set. This value represents
the average error that can be expected for the prediction of future
samples, with a confidence interval of 95