Resumen de Towards a global method for comparing classifiers

Alejandra Cabaña , Fabrice Gamboa

• It is well known, from empirical comparisons, that di erent algorithms show di erent performances when applied to di erent data sets. Our aim is to provide an universal instrument for choosing among I classi ers, when nothing is known a priori about the structure of data set to be classi ed. In this case, it might not suce to look just at classi cation mean errors: it would be advisable to use a classi cation algorithm with low variance.