Know more

 The evaluation of the developed models can be done at two levels: in considering the differences in predictions between ordinal observed and simulated classes , or by combining a quantitative value to the ordinal classes considered and using the quantitative criteria of normal characterization of the predictive performance of the models (bias, root mean square error of prediction and efficiency ) (Robin et al., 2013).
To complement these criterion, the work of Alan Agresti (2010) on the analysis of data ordinal categorical allow to implement different statistical procedures to characterize the degree of association between variables predicted and observed.
Different statistical criteria have thus been used: ....

Agresti A. 2010. Matched-pairs data with ordered categories. In: Analysis of ordinal categorical data. Second edition. Wiley& Sons, Inc. Hoboken, New-Jersey, USA. 225–261.