Sensors in the transportation domain, logs of electrical devices, and social media channels, among others, generate vast continuous streams of information that can be exploited with machine learning techniques, in order to design automatic systems that help taking better decisions. 

In the above situations, the challenge is not only the volume of the data, but also the fact that the data possesses complex spatial and temporal interconnections. Our research on spatio-temporal data modelling addresses these challenges in the following fields of application:

  • Transportation data
    : We study predicting demand and modelling user choices in public transport and parking operations, using historical data from public transport usage and on-street parking sensors.
  • Energy consumption: We measure and model the energy consumption profiles of electrical devices using statistical sequential models in order to optimize energetic performance. 
  • Social media: Our research applies machine learning techniques to social media data with the aim of filtering messages by category, and discovering static patterns or dynamic trends. 
  • Simulation: In the context of Urban Mobility, we are developing new capabilities to simulate the behaviour of different actors in a transportation network for the purpose of network diagnosis, optimization and what-if analysis. We use historical data and machine learning techniques to model vehicles in the network and passenger (multi-modal) plans; deploying these models in the simulator allows to more accurately reproducing the real traffic conditions comparing to ones based on assumptions from transport engineering.