The current model relies on trial and error iterations to calibrate the model, is there a more efficient way to improve efficiency