Predictive modeling of spatiotemporal phenomena in GIS using Machine Learning

This research focuses on modelling spatio-temporal phenomena from sensor-network data. This research originated in the quest to model pedestrian behaviour, crowdedness and crowd flows at mass-events, based on data from a network of Bluetooth-enabled sensors. These sensors use the amount of nearby Bluetooth-enabled cellphones as proxy for crowdedness (a technology developed at the CartoGIS research group of the department of Geography). Currently, the project focusses on modeling general spatio-temporal phenomena (e.g. traffic jams, air pollution, sea temperatures, etc.) from data collected using apposite sensor networks. It is investigated how Machine Learning techniques such as Graphical Models (Bayesian Networks, Markov Random Fields) can be used to learn the temporal and spatial relationships of such complex phenomena, and how this knowledge can be used to make predictions on the phenomenon's behaviour.



Research units



Staff members