Title: Bayesian spatio-temporal clustering for areal data
Authors: Annalisa Cadonna - WU, Vienna University of Economics and Business (Austria) [presenting]
Alessandra Guglielmi - Politecnico di Milano (Italy)
Andrea Cremaschi - Universitetet i Oslo (Norway)
Abstract: Recent availability of mobile data provides researchers with datasets which are collected over time and on a huge spatial grid. The goal is the development of appropriate models and efficient algorithms for clustering of large to huge spatio-temporal datasets. Specific focus is placed on their potential application to clustering regions based on population density dynamics. In fact, large scale quantitative information on population density dynamics is of great interest to urban planners and city managers. We analyze high-dimensional areal data describing the use over time of the mobile-phone network in this area. The goal is to identify and cluster sub-regions of the metropolitan area of Milan which share similar characteristics along time in terms of population density dynamics, and to allow for the clustering to vary over time. Moreover, we would like to be able to detect isolated activities taking place in specific locations and times within the metropolitan area. To reach our goal, we perform Bayesian spatio-temporal clustering using a non-parametric approach based on a time-varying Dirichlet process. Preliminary results show time varying clustering, interpretable in terms of population density dynamic, such as weekly daily work activities, commuting, and big isolated events.