Discrepancy Analysis of Activity Sequences

Abstract

Sequence alignment, also known as optimal matching, has recently received new attention for use in analysis of activity patterns. The method is almost always combined with a data reduction technique, such as clustering analysis. The cluster-based approach is powerful for discovery of a typology of activity patterns of people. However, the use of the combination of the sequence alignment and cluster analysis methodologies does not seem to be successful for the identification of diverse factors that would affect activity sequence patterns. This outcome is because the loss of too much information may occur when the set of activity sequences is reduced to a small number of clusters. This paper proposes the use of a new combination of the sequence alignment and discrepancy analysis methodologies instead of the cluster-based approach. As a generalization of the principle of analysis of variance, discrepancy analysis allows the association between activity sequences characterized by a pairwise distance matrix and one or more covariates to be evaluated. In addition, an induction tree complements the sequence discrepancy analysis and displays how individual activity sequences vary with the value of covariates.