Meta Learning Application in Rank Aggregation Feature Selection

One of the main tasks of machine learning and data mining is feature selection. Depending on the task different methods applied to find optimal balance between speed and feature selection quality. MeLiF algorithm effectively solves feature selection problem by building ensemble of feature ranking filters. It reduces filters aggregation problem to linear form optimization problem and works as a wrapper, but not on feature space as classical wrappers do, but on linear form coefficients space, which is much smaller. In this paper we tried to apply meta-learning to provide good starting optimization points for MeLiF method and as a result we increased not only speed but in some cases feature selection quality of this method.