neural additive models (nams)

These models combine some of the benefits of neural networks and decision trees. They are designed to extrapolate well and are interpretable.