Abstract
Sequential pattern mining is a key technique of data mining with broad applications (user behavior analysis, bioinformatic, security, music, etc.). Sequential pattern mining aims at discovering correlations among events through time. There exist many algorithms to discover such patterns. However, these approaches only take one dimension into account (e.g. product dimension in customer market basket problem analysis) whereas data are multidimensional in nature. In this thesis, we define multidimensional sequential patterns to take the specificity of multidimensional databases (several dimensions, hierarchies, aggregated value). We define algorithms that allow the discovery of such patterns by handling this specificity. Some experiments on both synthetic and real data are reported and show the interest of our proposals. We also focus on the discovery of atypical behavior. We show that there are several interpretations of an atypical behavior (fact or knowledge). According to each interpretation, we propose an approach to discover such behaviors. These approaches are also validated with experiments on real data.