| Area | Detail |
|---|---|
| Dataset | City of Austin construction permits |
| Tools | Python, Pandas, Scikit-learn, Matplotlib |
| Focus | ZIP-level construction intensity and added square footage modeling |
| Target | TotalNewAddSQFT |
| Model | Random forest regression |
| Best use case | Urban growth analysis and infrastructure planning signals |
| Caution | Predictive, not causal; not a production forecast |
This project analyzes City of Austin construction permit data to identify where new development is concentrated and to test whether permit-level features can help model added square footage.
The project is best understood as a descriptive analysis plus predictive prototype. It uses construction permits as a faster-updating signal than census data, while treating model results cautiously.
This map visualizes where construction permit activity is concentrated across Austin, helping identify areas with stronger development pressure.
Where is construction activity concentrated in Austin, and can permit metadata help predict added square footage as a proxy for development intensity?
- Python
- pandas
- matplotlib
- scikit-learn
- Jupyter Notebook
Source: City of Austin Open Data Portal
Original project scope: more than 1.8 million construction permit records from 2010–2017
Working analytical subset: about 241,000 records after filtering to rows with positive added square footage
Key fields used:
TotalNewAddSQFTOriginalZipLatitudeLongitudePermitClassStatusCurrentCouncilDistrictNumberOfFloorsHousingUnitsTotalJobValuation
Austin's growth affects housing, utilities, transportation, and infrastructure planning. This project asks:
- Which Austin ZIP codes show the greatest concentration of new construction?
- Which permit-level features are most useful for modeling added square footage?
This is not a causal analysis. It does not prove that permits cause infrastructure demand. It identifies patterns in construction activity.
The analysis filters permit records to rows with positive TotalNewAddSQFT, coerce numeric fields, normalize valuation fields, and remove records missing key geographic fields needed for ZIP-level and map-based analysis.
EDA focuses on:
- permit counts by ZIP code
- added square footage by ZIP code
- permit counts by number of floors
- most common permit classes
- job valuation distribution
- geographic distribution of added square footage
The modeling script treats TotalNewAddSQFT as the target variable and uses a random forest regressor with permit-level metadata.
The pipeline includes:
- numeric imputation
- categorical imputation
- one-hot encoding
- random forest regression
- grid search cross-validation
- train/test evaluation with MAE and R²
The project currently reports the following high-construction ZIP codes by added square footage:
| ZIP Code | Reported Added Square Footage |
|---|---|
| 78701 | 106 million |
| 78704 | 89 million |
| 78758 | 75 million |
| 78748 | 70 million |
| 78744 | 60 million |
The project also reports that location-based fields such as ZIP code, latitude, and longitude are informative, with NumberOfFloors also contributing a useful signal.
The README previously reported:
- random forest R² of 0.72
- monthly time-series split R² of 0.66
Those values suggest a useful predictive signal, but they should be treated as provisional until the full validation strategy is documented end-to-end in reproducible scripts.
Construction permits update more frequently than census releases. That makes them useful for detecting neighborhood-level development patterns earlier than slower demographic datasets.
In this project, added square footage is used as a rough proxy for construction intensity and possible downstream infrastructure pressure.
This project demonstrates:
- Python EDA
- civic/open-data analysis
- feature engineering
- regression modeling
- cautious interpretation of model performance
- communication of limitations