As many as 16-18 factors were needed for optimal model
performance. When the three waxy samples of the calibration
set and the two of the validation set were omitted along with
the Texas-grown amylose extender sample of the calibration set,
the r2 value of the validation set declined to 0.94, although the
SEP remained virtually unchanged. The effect of the second-difference
pretreatment was to reduce the number of factors required
for optimal performance by two. The standard errors at optimal
conditions were equivalent between the standard and seconddifference
pretreatments.