Ordinal Prototype-Based Classifiers

The identification of prototypical patterns is one of the major goals in the classification of microarray data. Prototype-based classifiers are of special interest in this context, since they allow a direct biological interpretation. In this work we present prototype-based classifiers that rely on ordinal-scaled data. Advantage of these ordinal-scaled signatures is their invariance to a wide range of data transformations. Standard prototype-based classifiers can be modified to this type of data by utilizing rank-distances and rank-aggregation procedures. In this study, we compare the proposed methods with standard classifiers. They are examined in experiments with and without feature selection on a panel of publicly available microarray datasets. We show that the proposed techniques result in the construction of different signatures that improve classification performance.