An Efficient Interval-Based Approach to Mining Frequent Patterns in a Time Series Database

In this paper, we introduce an interval-based approach to mining frequent patterns in a time series database. As compared to frequent patterns in the existing approaches, frequent patterns in our approach are more informative with explicit time gaps automatically discovered along with the temporal relationships between the components in each pattern. In addition, our interval-based frequent pattern mining algorithm on time series databases, called IFPATS, is more efficient with a single database scan and a looking-ahead mechanism for a reduction in non-potential candidates for frequent patterns. Experimental results have been conducted and have confirmed that our IFPATS algorithm outperforms both the existing interval-based algorithm on sequential databases and the straightforward approach with post processing for explicit time gaps in the temporal relationships of the resulting patterns. Especially as a time series database gets larger and time series get longer in a higher dimensional space, our approach is much more efficient.