A meta-learning framework for pattern classification by means of data complexity measures
José Martínez Sotoca, Ramón A. Mollineda, J. Salvador Sánchez · INTELIGENCIA ARTIFICIAL · 2006
It is widely accepted that the empirical behavior of classiflers strongly depends on available data. For a given problem, it is rather di-cult to guess which classifler will provide the best performance or to set a proper expectation on classiflcation performance. Traditional experimental studies consist of presenting accuracy of a set of classiflers on a small number of problems, without analyzing why a classifler outperforms other classiflcation algorithms. Recently, some researchers have tried to characterize data complexity and relate it to classifler performance. In this paper, we present a general meta-learning framework based on a number of data complexity measures. We also discuss the applicability of this method to several problems in pattern analysis.