Bootstrap ez-parametrikoa algoritmoen errendimenduaren konparaketarako estatistika bayestarraren alternatiba gisa

Abstract

In times where artificial intelligence is being incorporated into almost all application domains, it is essential to make comparisons between the results of the algorithms we use. In these comparisons, the quantification of the uncertainty of the results obtained is an important aspect, but it is generally not given sufficient importance. In fact, the frequentist statistical tests that have been used over the years have obvious shortcomings in this regard. As an alternative, Bayesian statistics is the most widely accepted solution today, but there are other methods that have not yet been explored in depth. In this article we will examine the use of non-parametric bootstrap to illustrate this uncertainty, and we will compare the behavior of this method with that of Bayesian statistics, underlining the similarities and differences.