>If you think the model and it's assumptions are wrong why not go build a better model?
The better model may be intractable to fit (so complex that we lack enough data to train it). Or, it may be fittable but not easy to understand, e.g. a neural network model.
This. I work with models that are considerably simpler than "200 equations and 150 parameters" and we run into non-identifiability problems all the time.
The better model may be intractable to fit (so complex that we lack enough data to train it). Or, it may be fittable but not easy to understand, e.g. a neural network model.