There is no free lunch. "Best" depends on what you want to use it for.
The ones you mention are quite different from each other; supervised, non-supervised... So please think about one task, and then check out what kind of neural networks are used for it.
If you don't want to use them, I guess them "best" is would be the one you find the most papers about. Which would probably be Support Vector Machines, probably. But my point was that you can't talk about "best" if you don't add a predicate: best for something or other.