Analysis of E-Scooter Crash Severity Using Interpretable Machine Learning Approaches

Please look at the website of the host laboratory SATIE by https://satie.ens-paris-saclay.fr/fr/presentation

and the website of UGE by https://fr.wikipedia.org/wiki/Universit%C3%A9_Gustave-Eiffel

E-scooters have become an integral part of urban mobility systems in large metropolitan areas. Understanding the determinants of e-scooter crash severity is therefore crucial for improving road safety, guiding infrastructure design, and supporting evidence-based urban transport policies. However, crash severity is driven by complex and nonlinear interactions between user profiles, crash characteristics, and environmental conditions. Traditional statistical approaches often struggle to adequately capture this heterogeneity.

Supported by our AMI-QIM project, the objective of this internship is to analyze e-scooter crash severity in the Greater Paris area using interpretable machine learning approaches, such as gradient boosting algorithms (e.g., XGBoost or LightGBM) and SHapley Additive exPlanations (SHAP). The study will use the French road traffic accident database “Observatoire National Interministériel de la Sécurité Routière (ONISR)” for the years of 2022–2024.

This internship is expected to identify distinct e-scooter crash typologies in the Greater Paris and to improve understanding of heterogeneous severity determinants from user socio-demographics (e.g., age, gender), environmental and infrastructural conditions (e.g., road type, intersection presence, weather), and accidental situations (e.g., collision type, vehicles involved). The study results are expected to provide policy-relevant insights to support targeted road safety measures.

Reference

1) Pervez, A. and Jamal, A. (2025). Exploring e-scooter risk factors based on interpretable machine learning framework. Journal of Safety Research, 94, 128–140.

2) Abdi, A. and O’Hern, S. (2025). Understanding e-scooter rider crash severity using a built environment typology: A two-stage clustering and random parameter model analysis. Accident Analysis & Prevention, 215, 108018.

3) Md Monzurul Islam et al. (2026). A multidimensional analysis of E-scooter crash severity: Integrating cluster correspondence and SHAP interpretability. The 105th TRB Annual Meeting, Washington, D.C. USA.

We are looking for a motivated M2 (Master’s level) student in data science, artificial intelligence, or a related field, with a strong interest in road safety and urban mobility. The ideal candidate should demonstrate: (1) A solid background in data analysis and statistics, including experience with multivariate data, and practical knowledge of machine learning methods; (2) Programming skills in Python or R, with experience using relevant data science libraries; (3) An interest in interpretable and explainable AI methods, or a strong motivation to learn tools such as SHAP.