Investigating The Factors Affecting Cycling Accident Severity: A Bayesian Hierarchical Mixtual Model Apporach

We developed a Bayesian hierarchical mixture model using PyMC to investigate factors influencing cycling accident severity using over 800,000 records from the UK’s open‑source accident database. The project combined extensive data cleaning, missing‑value imputation via posterior predictive sampling, and exploratory analysis in Python to select meaningful predictors for downstream modeling. In the model, we included hierarchical time‑of‑day structures, spike‑and‑slab variable selection, and demographic mixture components (gender and age), achieving improved model fit based on WAIC/ELPD comparisons. The final model provided interpretable insights into how environmental conditions, lighting, and cyclist characteristics shape the likelihood of severe injuries.

Click slides and presentation videoto view more.