Integrating association rule mining algorithms with the F2 OODBMS

Abstract

In this paper, we integrate two association rule mining algorithms, Apriori and TRAND, with the F2 object-oriented database system (DBMS). The advantages of our integration are the following. Both algorithms do not need to maintain complicated data structures and use only database classes. Both algorithms do not need to manage the buffer since it is handled by the DBMS. Both algorithms store frequent itemsets in the database which can be retrieved later using the DBMS data manipulation language. In addition to that, the TRAND algorithm takes advantage of the transposed storage supported in F2. To compute the support of candidate itemsets, it applies logical AND on boolean attributes, implemented in F2 as vectors, and avoids scanning the database. This reduces significantly the number of block accesses and consequently the execution time. © Springer-Verlag Berlin Heidelberg 2003.