Selecting prototypes for two multicriteria classification methods: A comparative study

In the last years, the area of Multicriteria Decision Analysis (MCDA) has brought about new methods to cope with classification problems, among which those based on the concept of prototypes. These refer to specific alternatives (samples) of the training dataset that are good representatives of the groups they fit in. In this paper, experiments are conducted over two prototype selection (PS) techniques employed to improve the accuracy of two prototype-based MCDA classification methods. The PS techiques investigated are based, respectively, on a customized genetic algorithm and on the Electre IV approach, whereas the MCDA classification methods studied comprise the one proposed by Goletsis et al. and the well-known PROAFTN method. The results achieved demonstrate that the classification methods are indeed very sensitive to the choice of prototypes and that the PS techniques investigated may be instrumental for leveraging their performance levels.