Estimating the predictive accuracy of classifiers and ranking them

Abstract

This paper investigates the use of meta-learning to estimate the predictive accuracy of a classiier. We present a scenario where meta-learning is seen as a regression task and consider its potential in connec-tion with three strategies of dataset characterization. We show t h a t it is possible to estimate classiier performance with a high degree of con-and gain knowledge about the classiier through the regression models generated. We also show that the best strategy for performance estimation is not necessarily the best one for model selection.