Heinrich-Heine-Universität Düsseldorf

Market basket analysis is one important application of knowledge discovery in databases. Real life market basket databases usually contain temporal coherences, which cannot be captured by means of standard association rule mining. Thus there is a need for developing algorithms, that reveal such temporal coherences within this data. This paper gathers several notions of temporal association rules and presents an approach for mining most of these kinds (cyclic, lifespan- and calendar-based) in a market basket database, enhanced by two novel tree structures. We called these two tree structures EP- and ET-Tree, which are derived from existing approaches improving standard association rule mining. They are used as representation of the database and thus make the discovery of temporal association rules very efficient.