On the use of meta-learning for instance selection: An architecture and an experimental study ☆

摘要

Many authors agree that, when applying instance selection to a data set, it would be useful to characterize the data set in order to choose the most suitable selection criterion. Based on this hypothesis, we propose an architecture for knowledge-based instance selection (KBIS) systems. It uses meta-learning to select the best suited instance selection method for each specific database, among several methods available. We carried out a study in order to verify whether this architecture can outperform the individual methods. Two different versions of a KBIS system based on our architecture, each using a different learner, were instantiated. They were evaluated experimentally and the results were compared to those of the individual methods used.