[ely-s]

This is interesting experimental evidence in spite of the NFL theorem, which refutes the notion of a generally superior algorithm.

https://en.wikipedia.org/wiki/No_free_lunch_theorem

I would reconcile it by saying that the UCI contains a biased subset of all theoretical classification tasks.

[Houshalter]

This is not surprising to anyone, because the no free lunch theorem assumes that all datasets are randomly generated. In reality real world problems are probably drawn from some distribution, and datasets are not totally random.