UrbanFlow Valencia

City teams need both reliable hourly traffic forecasts and a defensible way to allocate facilities under budget constraints.

The project separates prediction from decision: hourly CatBoost models forecast traffic, while a PuLP/CBC optimizer selects facility locations independently.

FastAPI, Next.js, Docker and GitHub Actions publish the system, while monitoring computes MAE by hour and flags systematic errors by zone.

The model is trained on one month of traffic data; seasonal generalization requires additional validation.

• UrbanFlow Valencia repository
  README — model, evaluation and deployment