Well, you can kind of use decision forests, and similar approaches, to combine algorithms. The lack of human-readability makes it pretty tough unless you're in a domain where you can effectively apply weighted heuristics (which I found to work pretty well in NLP, albeit with a lot of experimental tweaking to arrive at the right model for my domain).