Temporal Pattern Mining Using a Time Ontology

The analysis of temporal data has deserved a considerable attention, in particular in the analysis of time series. However, the general research on data mining seldom has focused its attention on dealing with the specific attribute – time. The discovery of temporal patterns, that reveal interesting behaviors over time, is one of such cases. In this paper, we propose a new approach to effectively find frequent temporal patterns. Our approach is based on the interleaved algorithm and the use of a time ontology, which precisely defines the main temporal concepts. Based on these concepts is then possible to effectively generate all possible time intervals when frequent patterns could occur, and then count each corresponding support, identifying frequent patterns. The algorithm follows the candidate generation and test philosophy, being able to deal with different time granularities.