Discovering quasi-periodic-frequent patterns in transactional databases

Abstract

Periodic-frequent patterns are an important class of user-interest-based frequent patterns that exist in a transactional database. A frequent pattern can be said periodic-frequent if it appears periodically throughout the database. We have observed that it is difficult to mine periodic-frequent patterns in very large databases. The reason is that the occurrence behavior of the patterns can vary over a period of time causing periodically occurring patterns to be non-periodic and/or vice-versa. We call this problem as the "intermittence problem." Furthermore, in some of the real-world applications, the users may be interested in only those frequent patterns that might have appeared almost periodically throughout the database. With this motivation, we relax the constraint that a pattern must appear periodically throughout the database, and introduce a new class of user-interestbased frequent patterns, called quasi-periodic-frequent patterns. Informally, a frequent pattern is said to be quasi-periodic-frequent if most of its occurrences are periodic in a database.We propose a model and a pattern-growth algorithm to discover these patterns. The proposed patterns do not satisfy the downward closure property. We have introduced three pruning techniques to reduce the computational cost of mining the patterns. Experimental results show that the proposed patterns can provide useful information and the proposed algorithm is efficient. © Springer International Publishing Switzerland 2013.