Abstract
In this paper a model of the classification task, based on the Monte Carlo theory of stochastic algorithms, is presented. The model allows different classification paradigms to be described, such as the single classifier, the ensemble, and the crowdsourcing ones. Some experimental work validates the theoretical predictions about the relatioship between the mean and variance of the classification accuracy and the number of classifiers.