Mining inter-transaction recurring patterns in time series

Mining recurring patterns plays an important role in finding regularities for particular time intervals within a time series. Most existing methods on recurring pattern search typically focus on finding the patterns that exhibit temporal associations among items within a transaction (timestamp). However, present methods for mining recurring patterns are not expressive enough to capture recurring events across multiple timestamps, which often occurs in many real-world applications. In this study, we extend the single-transaction mining scheme of recurring patterns to address the limitation. Our proposed method basically consists of two major steps. The first step is to construct inter-transaction database from an original database. The second step is to efficiently mine recurring patterns on the inter-transaction database and named inter-transaction recurring patterns. Experimental results conducted on a real-world database confirm that recurring patterns by our proposed method capture meaningful recurrences across multiple timestamps within a time series.