On April 23, 2026,Cédric Roussel successfully defended his dissertation titled *A Framework for Explainable Artificial Intelligence in Geospatial Contexts *. The thesis was written as part of the TOPML project and supervised at the Faculty of Engineering at the University of Freiburg. We extend our warmest congratulations to Cédric on this great achievement!
Klaus Böhm and Cédric Roussel (photo: Ulrike Roussel); doctoral cap (photo: Kira Zschiesche), CC BY-SA 4.0
In his dissertation, he investigated the transparency of complex machine learning models in a spatial context. The goal was to develop new methods and gain insights to make these systems—also known as black-box models—more understandable to humans. This falls under the research field of Explainable Artificial Intelligence ( XAI).
In addition to the well-known global and local explanations in XAI, Cédric Roussel developed a novel method based on so-called glocal explanations. These explanations bridge the gap between the two well-known extremes and, by aggregating local explanations, allow for a focus on defined spatial factors. Roussel visualized the newly developed methods and results in two new approaches, one of which is a geovisualization. A user evaluation demonstrated the clear advantage of this visualization compared to simple tables of results.
To apply and validate the developed methods in practice, Cédric Roussel used three real-world use cases. In Hamburg, he was able to use his method to predict booking numbers at approximately 200 bike-sharing stations and analyze user behavior. He was able to apply the same procedure to ten parking garages in Mainz. The third use case involved developing a model that determines the severity of traffic accidents. Roussel then investigated which factors contribute to a traffic accident resulting in more serious consequences, with the aim of enabling preventive measures.
The dissertation makes an important contribution to improving the transparency of complex machine learning models and provides new methods that are specifically suited for use cases with a spatial context.
The thesis was supervised by Prof. Dr. Alexander Reiterer (primary advisor and first reviewer) and Prof. Dr. Klaus Böhm (secondary advisor and third reviewer), and was additionally reviewed by Prof. Dr. Nina Hubig (secondary reviewer) in addition to the advisors. The doctoral proceedings were overseen by Prof. Dr. Ulrike Wallrabe (Chair of the Committee) and Prof. Dr. Daniel Carl (Member of the Committee).
