Quite clearly, the number of evaluations can't be the same when the search space is much bigger (because you're adding new neurons and layers). That is the reason why results for spambase get worse with several layers. So we need to fine-tune the number of evaluations (population/generations) in order to obtain better results.
Quite clearly, the number of evaluations can't be the same when the search space is much bigger (because you're adding new neurons and layers). That is the reason why results for spambase get worse with several layers. So we need to fine-tune the number of evaluations (population/generations) in order to obtain better results.