A1524
Title: Mixture of state space models for compositional data with an application to urban mobility analysis
Authors: Andrea Panarotto - Department of Statistical Sciences, University of Padova (Italy) [presenting]
Manuela Cattelan - University of Padova (Italy)
Ruggero Bellio - University of Udine (Italy)
Abstract: Capturing the dependency between successive compositional observations requires models that account for the constrained nature of the data. A framework is introduced for the analysis and clustering of compositional time series, where the trajectories evolve within the simplex. The approach integrates a state-space representation of compositional dynamics into a model-based clustering framework, enabling the identification of groups of trajectories sharing similar temporal patterns. The model can be extended through a mixture of experts model, allowing trajectory-level covariates to influence the component weights. The methodology is motivated by an application to urban mobility, where individuals' movements are represented in the simplex by the proportions of types of roads in their surroundings. This formulation provides a data-driven way to aggregate individual travel behaviors into population-wide mobility patterns, offering new insights into how people interact with different urban environments over time.
