Spatio-temporal data Analysis and Reasoning (STAR)

GROUP AIM

Our research group is part of the Automated Design of Algorithms (ADA) embedded in the Leiden Institute of Advanced Computer Sciences (LIACS) at Leiden University. We aim to design algorithms for effectively processing spatial, time-series and spatio-temporal data. In principle, every process around us is spatio-temporal, as we can attach time and space to it. Spatio-temporal datasets represent the development of these processes by sampling them in time and space. Modern sensing technologies such as Earth Observation satellites, GPS, or wearables sensors have allowed the collecting such datasets at large scales. We explore the design of algorithms that can automatically handle all necessary data processing tasks from the point of data collection to high-level modelling, extraction of information, and effective decision-making from such data. Our research targets complex applications in a broad range of urban, environmental, and industrial domains. We strongly focus on developing algorithms for wearables data, Earth observations and open spatio-temporal data sources. Find more about our new initiative AutoAI4EO, where we push for advancing machine learning algorithms for Earth observations.

NEWS

Jan 15, 2026

New visitor
Jurairat Preechasin is vising the STAR Research Group from 15 January–15 April 2026.

Jan 10, 2026

PhD thesis submitted
PhD thesis of Laurens Arp is submitted to the committee!

Nov 8, 2025

Paper accepted
Erfan Moeini has a paper accepted to AAAI 2026 titled Neural Architecture and Hyperparameter Selection through Meta-Learning on Time Series.

Nov 1, 2025

New PhD student started
Sina Ghorbani Kolahi started as an adjunct PhD student on the NWA project Breaking the cycle.

Sep 19, 2025

Best paper award
Julia Wasala was awarded a best paper award at MCLEAN-ECMLPKDD workshop.

RESEARCH

A screenshot of the portfolio page

Since 2011

Trajectory data
Spatio-temporal trajectories show the movement of moving objects (e.g., humans, cars, animals) over space. We design algorithms that can extract patterns from such data. Human Trajectory data can be used to unravel information about social dynamics and space usage. Within an earlier project funded by the Center for BOLD citites we looked at trajectories of pupils playing in the playgrounds. In a newly funded project by NWO we are taking a step further by connecting such information with features of the built environment and physiological indicators. By doing so, individual differences in the subjective experience of loneliness are linked to the context in which they arise, enabling the identification of intervention strategies and testing.

Since 2023

Wearables data
Within the LABDA (Learning Network for Advanced Behavioural Data Analysis) EU project, we would like to understand how 24/7 observational activity data collected by wearable sensors (e.g., smart watches) can be used to identify and recommend effective changes in daily activities (i.e., possible behavioural interventions) that are expected to result in concrete health improvements.
A screenshot of the portfolio page

Since 2022

Astronomical observations
 We study how machine learning algorithms developed for data acquired from telescopes can help understand the world better. For this goal, we strongly collaborate with astronomers in the Leiden Observatory.
A screenshot of the portfolio page

Since 2017

Urban data
We would like to know how open and ubiquitous data sources can help find computational solutions for urban challenges. Within the master's course Urban Computing, we cover methods for processing spatio-temporal data. Students complete course projects that address urban problems using computational methods. Notably, in 2019, three of our projects made it up to the top 10 selected contributions in the Future Cities Challenge.
A screenshot of the portfolio page

Since 2018

Time series
Time series data are observations of a single process of regular time intervals. We have previously designed different algorithms for time series data for different applications (e.g., energy load forecasting, remaining useful life estimation, and COVID-19 forecasting).

RECENT PUBLICATIONS

[All Publications]
  1. Neural Architecture and Hyperparameter Selection through Meta-Learning on Time Series
    Erfan Moeini, Christopher Vox, Marie Anastacio, Wadie Skaf, Mitra Baratchi, and Holger H. Hoos
    In Proceedings of the AAAI Conference on Artificial Intelligence (To appear), 2026
  2. Best Practices For Empirical Meta-Algorithmic Research Guidelines from the COSEAL Research Network
    Theresa Eimer, Lennart Schäpermeier, André Biedenkapp, Alexander Tornede, Lars Kotthoff, Pieter Leyman, Matthias Feurer, Katharina Eggensperger, Kaitlin Maile, Tanja Tornede, and others
    arXiv preprint arXiv:2512.16491, 2025
  3. Time Series Representations Classroom (TSRC): A Teacher-Student-based Framework for Interpretability-enhanced Unsupervised Time Series Representation Learning
    Wadie Skaf, Mitra Baratchi, and Holger Hoos
    Machine Learning, 2025

CONTACT

Address: Einsteinweg 55, 2333 CC Leiden, The Netherlands
Email: m.baratchi[at]liacs.leidenuniv.nl