Possibilistic Methodology for the Evaluation of Classification Algorithms

In the paper we consider the problem of the statistical evaluation and comparison of different classification algorithms. For this purpose we apply the methodology of statistical tests for testing independence in the case the multinomial distribution. We propose to use two-sample tests for the comparison of different classification algorithms. In the paper we consider only the case of the supervised classification when an external ‘expert’ evaluates the correctness of classification. The results of the proposed statistical tests are interpreted using possibilistic methodology based on indices of dominance introduced by [7].