Relation restricted prediction analysis

Title:	Relation restricted prediction analysis
Authors:	Ivo Düntsch , Dept of Computer Science , Brock University , St Catherines, Ontario, L2S 3A1, Canada Günther Gediga , Institut für Evaluation und Marktanalysen; Brinkstr. 19; D-49143 Jeggen; Germany (Equal authorship implied)
Status:	Proceedings of the 15th IMACS World Congress, Berlin, Ed. A. Sydow (1997), 619 - 624
Abstract:	For the description of dependencies between a set of independent attributes Q and a dependent attribute p, a soft computing approach such as Rough Set Data Analysis (RSDA) uses only a very simple data representation model: The set of equivalence classes of feature vectors determined by Q and p respectively. Although this model is satisfactory for many applications, there are sometimes problems to interpret the results, because this type of prediction does not take into account relational information within the attributes, for example, orderings. We consider the problem what form prediction should take in the ``nominal attributes predict an ordinal attribute'' situation ((n,o) - prediction ), as well as in the (o,o)-situation. We show how to define rough (n,o)- and (o,o)-prediction and approximation in terms of relational compatibility, which respects the granularity information given by the attributes. A running example is presented to demonstrate the result of the three types of data analysis.