In all modeling techniques, and neural networks in particular, care must be taken not to overtrain or overfit the model. If possible, models should be interpreted in terms of their mechanistic meaning.