Comparative analysis of multidimensional sequential trajectories clustering methods

Overview & Synopsis

Multidimensional sequence analysis is a central method widely used in social science research to identify underlying relationships between two or more trajectories observed simultaneously. This paper presents an extensive review of the existing methods for trajectory analysis, characterized by discrete or categorical longitudinal data, and provides an overview of their applications and differences in social science. The study focuses on non-parametric sequential clustering approaches, and compare various similarity measures applied on multidimensional sequences for both hard and soft clustering methods. Ccost and Cdist, two similarity measures for multivariate sequences have proven to be the most effective, but they are based on major assumptions concerning the interdependence of dimensions.