Filtering in multicriteria discrete problems: An approach

Most real-world applications, specially selection problems, can be modeled as a Discrete Multicriteria Problem (DMP) and concisely expressed in matrix format. In recent years, several Decision Support Systems have been proposed to help the Decision Maker (DM) in the ranking of the alternatives or in the selection of the ‘best’ compromise solutions, regarding the problem. Problem size is, perhaps, the major common structural concern for the method, due to the consequent increase on the number of questions asked to the DM and computer response time. In order to minimize those disadvantages, we think that it is usually wise to begin with an exploratory examination of the data, to get a ‘feeling’ about the impact matrix. The main idea is to clarify the general structure of the data, filtering it and get, if possible, a ‘reduced’ matrix (lower dimensionally). Viewing each vector of criteria values as a realization of a multivariate random variable, it is possible to use various techniques of multivariate statistical analysis to illustrate the approach.