Discovering periodic patterns common to multiple sequences

Abstract

Discovering periodic itemsets in transaction databases is an emerging data mining task. However, current algorithms are designed to discover periodic itemsets in a single sequence. But in real-life, it is desirable to find periodic patterns that are common to multiple sequences. For example, a retail store manager can benefit from finding that many customers buy the same products every week in a retail store, to adapt its marketing and sale strategies. To address this drawback of previous work, this paper defines the problem of mining periodic patterns common to multiple sequences and proposes an efficient algorithm named MPFPS, which relies on a novel PFPS-list structure and two novel periodicity measures to assess periodicity with more flexibility. Experiments on several synthetic and real-life databases show that MPFPS is efficient and can filter many non-periodic itemsets to reveal the desired patterns.