These models don't make any assumption about the functional form of . They can fit a wider range of shapes for but require more data for an accurate fit and are more likely to overfit to training data. More flexibility in the model leads to a lower interpretability.