Self-organizing maps applied to clustering in MCDA
Résumé
In the field of Multi-Criteria Decision Aiding (MCDA) the problem of clustering, has received less attention than the classical problems. In this presentation, we continue our previous work on formally defining clustering in MCDA by extending the self-organizing maps of Kohonen to clustering in this context. When using measures of similarity, the K-means algorithm can in fact be identified as a particular case of the Kohonen maps, a case where the nodes in the map are completely disconnected from each other. We extend the Kohonen maps to clustering in MCDA, by firstly using measures from this domain in order to compare the alternatives together, and secondly proposing different map topologies in order to look for different types of structures over the clusters of alternatives.