The problem of clustering has been widely studied in the context of data mining, where by grouping objects that are similar and separating those that are dissimilar we are able to find the natural structure of the data, from which we can then extract knowledge by using different summarizing measures. The field of Multiple Criteria Decision Aid (MCDA) focuses on modeling the preferences of real decision-makers and aids them in reaching certain decisions. While problems like choice, sorting and ranking have been thoroughly studied in MCDA literature, the problem of clustering has received less attention. Furthermore, many clustering approaches in MCDA use measures related to similarity and do not exploit the additional preferential information that is present in this context. The thesis addresses these issues by first drawing a parallel between clustering in data mining and MCDA and redefining this problem in the latter field. Several different models are then proposed for the problem, as well as a few algorithms for solving it, which are then validated over a large set of benchmarks which have been created specifically for this purpose, by encapsulating as many and diverse potential problems inside them. Finally we consider a few practical applications through the use of several descriptive measures and extensions of the algorithms for handling large datasets, which are illustrated in part over a real case study.
