Up to now there is no research aiming to either predict the accuracy of ensemble
learning schemes or to predict which ensemble learning scheme to use for a given
dataset. Here we have investigated both tasks and found them to work quite
well. We give some representative examples.
Using regression to predict the performance of basic learning algorithms
was first investigated in (Gama & Brazdil, 1995), continuing the framework of
StatLOG(Brazdil, Gama & Henry, 1994). They report poor results in terms of
normalized mean squared error.