Why do some neighborhoods improve, while others deteriorate? While social scientists have long studied the question, recent advances in data collection and machine learning algorithms have made exploring the topic much less cumbersome. Using a computer-vision system, MIP network leader Scott Kominers and co-authors Nikhil Naik, Ramesh Raskar, Edward Glaeser, and César Hidalgo study which characteristics of a neighborhood predict improvement.

Out this month in the Proceedings of the National Academy of Sciences, the paper tracks changes in five U.S. cities using time-series street-level photographs taken in 2007 and 2014. They connect their observations on the physical appearance of the cities with economic and demographic data and find three factors that predict neighborhood improvement. Their findings show that education and population density can predict improvements in an area's physical environment, and that neighborhoods with better initial appearances tend to lead to greater positive improvements. Lastly, neighborhood improvement correlates positively with both the central business district and other physically attractive areas.

"Together, our results provide support for three classical theories of urban change and illustrate the value of using computer vision methods and street-level imagery to understand the physical dynamics of cities," the authors note.

You can read the paper, "Computer vision uncovers predictors of physical urban change," here. You can also view interactive maps of the areas studied here. This research was funded in part by HCEO.