In this project the main objective is to develop new AI methodology for explaining (uncertainty in) class predictions, with applications in the context of new mobility concepts.

Problem Context

Together with governmental partners, the aim of New Mobility is to zoom in on which factors in the region determine transport mode choices. The hypothesis is that these factors are not static but change due to socio-economic and technological developments. The objective of New Mobility is to test this hypothesis and subsequently to move towards continuous (regional – Brainport area - and/or provincial) monitoring of the decisive factors.


Explainability can help the governmental partners reach the above-mentioned objection. The explainability of the AI models provides more insights regarding the policy measures that are most effective for steering mobility. This is relevant for assessing and predicting the impact of policy measures on mode choice making it possible for the policy maker to investigate different ‘what if scenario’s’ and adapt the policy measures accordingly towards more effective and more supported by the population.


In 2021 we developed PERFEX, a model-agnostic method to generate explanations for the performance of a classifier. The performance can be assessed based on a user-defined metric. In 2022 we finalized this line of work by analyzing the relation with LIME (Ribeiro et al, 2016). The results of this analysis have been added to the draft paper that was written in 2021. The paper has been accepted for presentation in a workshop on Deployable AI, part of the AAAI-23 conference (Walraven et al, 2023).


New Mobility is partially funded by the province of Noord-Brabant. The project additionally collaborated with the municipalities of Eindhoven and Helmond, and with the organisation SmartwayZ.NL.


  • Taoufik Bakri, Mathematician, TNO, e-mail: