By analyzing historical data sequences and identifying relations between the occurring of data items and certain types of business events we have opportunities to gain insights into future status and thereby take action proactively. This paper proposes a new approach to cope with the problem of prediction on data sequence characterized by multiple dimensions. The proposed relation mining approach improves the existing sequential pattern mining algorithm by considering multidimensional data sequences and incorporating time constraints. We demonstrate that multidimensional relations extracted by our approach are an enhancement of single dimensional relations by showing significantly stronger prediction capability, despite of the substantial work done in the latter area. In addition, matching algorithm based on the obtained relations is proposed to make prediction. The effectiveness of the proposed methods is validated by experiments conducted on a mobile user context dataset.
Tang, Heng; Liao, Stephen Shaoyi; and Sun, Sherry Xiaoyun, "Mining Sequential Relations from Multidimensional Data Sequence for Prediction" (2008). ICIS 2008 Proceedings. 197.