PROJECT / 002
CrashCast
Spatiotemporal Traffic Collision Prediction
ILLUSTRATIVE / SYNTHETIC
01 / Overview
A geospatial modeling study of where and when traffic collision risk may vary.
02 / Problem
The challenge is to define a useful spatial unit and prediction window without allowing nearby or future information to leak into evaluation.
03 / Approach
The project explores spatial feature engineering and geographically aware validation for interpretable risk estimates.
04 / Technical stack
Python · GeoPandas · PostGIS · H3 · XGBoost
05 / Visualization
Illustrative visualization only. It communicates the project direction rather than a completed experimental result.
06 / Future development
The final evaluation will account for spatial separation and practical tradeoffs between signal, coverage, and interpretability.