Mining relational patterns from multiple relational tables

Abstract

In this paper, we present the concept of relational patterns and our approach to extract them from multiple relational tables. Relational patterns are analogous to frequent itemsets extracted by the Apriori algorithm [R. Agrawal, H. Mannila, R. Srikant, H. Toivonen, A.I. Verkamo, Advances in Knowledge Discovery and Data Mining, AAAI Press, 1995.] in the case of a single table. However, for the multiple relational tables, relational patterns capture co-occurrences of attributes as well as the relationships between these attributes, which are essential to avoid information loss. We describe our experiences from a test-bed implementation of our approach on a real hospital's discharge abstract database. This process raised issues, which were then implemented in order to enhance an analyst's ability to explore patterns while preventing high diversity and abundance of available data from blurring subtle patterns of interest. Finally, we evaluate the usefulness of relational patterns in the context of the discharge abstract data as well in other possible domains.