A Matter of Time: Multi-time Interval Pattern Discovery to Preserve the Temporal Information in between

Data Mining is concerned with the tasks of finding trends, patterns and relationships among patterns and many approaches have been proposed in this area. In particular, these tasks become extremely relevant when dealing with time-oriented data and information. However, especially in the field of so called sequence mining, most of the methods have a sequence of events as outcome, having neither any knowledge about the intervals between them nor about after how much time a particular pattern will reoccur. The goal of this thesis is to investigate how Temporal Data Mining can be adapted and used to analyze multivariate time-oriented data having multiple granularities in order to discover multi-time interval patterns, relations among data (previously unknown) as well as visually support this process. In detail, the approach we propose extends in particular the so called I-Apriori algorithm to non transactional databases and tries to provide a more general as well as more customizable way to find multi-time interval patterns in time oriented data. The proposed approach has been applied to data coming from a shopping mall and data coming from a flight company. The results outlined the presence of interesting behaviour in the data previously unknown and were used as basis for further analysis.