Hello,

I am trying to identify some models using the Design Of Experiments method. First I identify the main parameters from a two levels fractional plan. Then I do a full factorial plan with the main parameters to identify a more precise model and see if the second order terms are significant.

The model can be very precise on the test points if I add a lot of non statistically significant terms (pvalue>0.1). But it becomes not predictive between the tests points. I would like to know if there is an "official" way to identify the "best" model in order to be able to easily extend the model between the test points and to say "ok, now it is the best model I can find with my DOE approach" ?

Another question : is this possible not to be able to fit any polynomial model to some datas ? Are there some phenomena not easily "modelable" ?

Thanks for you answers.

Rémi

More Rémi Thanwerdas's questions See All
Similar questions and discussions