Background:
Motor vehicle crashes (MVCs) are a significant problem in the US. The US experiences about 40,000 MVC deaths per year (link), more than twice as many MVC deaths per capita as the average high-income country (link), not to mention 2 million injuries per year (link). Urban planning and traffic engineering can reduce traffic accident rate and severity through a variety of means, such as road geometry, speed reduction measures, separation of different modes of traffic, etc.

Question: What features of the built environment have the biggest impact on MVC rate and severity?

Datasets:
- Kaggle has a pretty large dataset on traffic accidents in the US: https://www.kaggle.com/datasets/sobhanmoosavi/us-accidents
- Besides the obvious features like time, location, and severity, the dataset has lots of other features, e.g. weather, nearby signage, nearby junctions, etc.
- The EPA has lots of data about the built environment, https://www.epa.gov/smartgrowth/smart-location-mapping
- Features run the gamut: density, land use, urban design, transit service, accessibility by transit, accessibility by car, demographics, employment, etc.

Challenges:
- The EPA datasets cover virtually every census block group in the nation. We might want to restrict ourselves to a more manageable geographic region.
- Both the Kaggle and EPA datasets have a LOT of feature variables, so we’ll need to carefully extract the most relevant features.