Hi @eldad_galili,
I think the problem is trivial : you're using different X's for the prediction of SubGroup between your training/validation set and your test set (your other dataset). See the numbers after Res 1, Res 2, Res 3, ... that do not match between the two tables :
It seems you have fit a functional model on your first dataset, extract the coefficients and fit a NN on these coefficients. On the other hand, you also have fit another functional model on the other dataset, and extracted the coefficients (but in this case, the X's are different, they do not come from the same exact model), and try to predict the subgroup by using the formula from previous NN fitting on the previous X's. So the NN can't predict anything on this test dataset since it hasn't learned from these "new" X's. Even if these models may be close (they come from the same functional model family, Symlet 20), the knots used in fitting these models may be different, hence the different X's naming and the prediction not possible.
What I would recommend is combining the two datasets together and either add a validation column to separate the training/validation set from the test set (the other dataset), or hide and exclude the rows corresponding to the other dataset / test set. Then launch the FDE with your validation column (or only the training rows not excluded), so that only the training set is used to fit the model. You can then extract the coefficients (same for all runs) and fit a NN only on your training/validation set. Then, save the formula for predicting subgroup in the datatable, and you'll be able to predict subgroup for the test data.
Solution from a very similar problem here : https://community.jmp.com/t5/Discussions/Functional-Data-Analysis-and-Classification-How-to-calculat...
It's very important to have the same functional model fitting if you want to use the coefficients from this model to predict other properties.
Hope this will help you,