Comparing classifiers for ordinal variables

To choose a single category of a qualitative variable using its predicted probability distribution is the final task to solve a classification problem. In this study, five predictive criteria are proposed and compared with the modal one, which is the standard criterion. The predictive performances are evaluated considering a set of indicators built from the resulting 3x3 confusion matrix. The data used are the decimal betting odds on the matches, transformed in probabilities of loss, draw and win of the home team, coming from the Kaggle European Soccer database, for seasons from 2008/2009 to 2015/2016 of the Italian League Serie A.