Seattle Demand Modeling
850,000 records of parking and weather, joined to answer whether rain changes where people park.
- Timeline
- Sept 2024 – Dec 2024 · Complete
- Role
- Solo
- Modality
- Tabular · Geospatial
- Evaluated on
- R² vs baseline · feature ablation
- Random Forest
- R²=0.86
- Records
- 850K+
- Occupancy on rainy days
- +25%
- Rain correlation
- r=0.78
The problem
Parking demand across a city is uneven and time-dependent, and the obvious hypothesis, that weather moves it, needs two datasets that do not naturally line up.
What I built
A machine-learning pipeline over 850,000+ records forecasting demand across five Seattle zones, integrating two independent source systems via API (Seattle Open Data and the Open-Meteo weather API) on aligned hourly timestamps. Automated cleaning and feature engineering derived rainy-hour flags, temperature bins, and weekday-versus-weekend patterns, cutting manual preprocessing time by ~35%.
How I knew it worked
Compared Linear Regression against Random Forest and reached R²=0.86, meaning the model accounts for 86% of demand variance. The finding worth having was behavioural: occupancy runs 25% higher on rainy days (r=0.78). Adding time-of-day and day-of-week features improved short-term prediction accuracy by 9%, which is the kind of increment that only shows up if you are measuring the same thing before and after.