Arbiter Meta-Learning with Dynamic Selection of Multiple Classifiers

In data mining, the selection of appropriate classifier to estimate some unknown attribute of a new instance has an essential role for the quality of results. Recently interesting approaches with parallel and distributed computing have been presented. In this paper we discuss an approach that uses classifiers trained on a number of data subsets in parallel as in the arbiter meta-learning technique. We suggest collecting the performance information of base classifiers and arbiters and the use of this information during the application phase to select the appropriate classifier dynamically. Despite of many open questions we are convinced that the dynamic selection approach suggested in this paper includes interesting characteristics that at least in some situations offer benefits in comparison with other meta-learning approaches. 1. Introduction Data mining is the process of finding previously unknown and potentially interesting patterns and relations in large databases [5]. A typical d..