Measuring Change in the Built Environment with Machine Learning Strategies: Problems and Prospects

• In Event: Using Computer Vision to Analyze Urban Images and Answer Crime & Place Questions

Abstract

Machine learning provides a unique way to study the built environment on a large scale. This approach, however, is not without limitations. The focus of this work is on the capabilities and limitations of machine learning models’ ability to measure change in the built environment. Specifically, this work focuses on the use of the DeepLab semantic segmentation model as well as historical Google street view, using these tools to measure changes in buildings, fencing, vegetation, and other features of the built environment between 2008 and 2023 in San Diego, California. The goal is to assess the accuracy and usefulness of this approach. Next steps in this research will include incorporating additional models, examining whether features of the built environment are related to crime, and examining how change in the built environment affects crime at the street segment level.

Authors

• Cheyenne Hodgen, University of California, Irvine

• John R. Hipp, University of California, Irvine