Evolutionary optimization of meta data metric for method recommendation

Metalearning - a method for recommendation the most suitable data-mining algorithm to an unknown dataset - is an important problem that needs to be solved in order to design a completely autonomous data-mining solver. This paper deals with this particular problem by proposing a machine-learning method which recommends the most suitable algorithm to an unknown dataset based on the results of previous data-mining experiments. The fundamental idea behind this is that the algorithms will perform similarly on similar datasets. The choice of datasets features - called meta data - is presented and the metric comparing datasets is optimized by means of evolutionary computation.