Crash typology paper published in DST

“A Roadway Crash Typology of Census Tracts Enables Targeted Interventions via Interpretable Machine Learning” is out in Data Science for Transportation (Vol. 7, Article 14). The paper clusters 2,480 New England census tracts into distinct crash profiles using UMAP and GMMs, then builds gradient boosting models with SHAP to explain which roadway and demographic factors drive crash risk in each cluster — so safety interventions can actually be targeted instead of one-size-fits-all.