Figure 5: Example to demonstrate
how many fractions and how many runs are required to train a statical
model in chromatography. Modeling error as a function of number of
fractions per run (n f) for values of coefficient
of variation (cv ) representing accuracy of analytical method. (B)
Pseudo-3D contour plot with prediction error as a function of bothn r and n f. From (Felfödi
et al., 2020)
Interpolation versus extrapolation
When setting up a prediction model the design space of critical process
parameters has to be considered.