Connecting state and local government leaders

Predicting Obesity From a City's Infrastructure

The algorithm examined Google Maps images from six cities, including Seattle, Tacoma and Bellevue, Washington.

The algorithm examined Google Maps images from six cities, including Seattle, Tacoma and Bellevue, Washington. Shutterstock


Connecting state and local government leaders

An artificial intelligence algorithm found a correlation between green space and obesity in several cities.

It may be possible to determine the prevalence of obesity in cities by analyzing infrastructure rather than residents, according to a recent study by researchers from the University of Washington.

The duo, Adyasha Maharana and Elaine Okanyene Nsoesie, created an artificial intelligence algorithm that examined roughly 150,000 high-resolution satellite images from Google Maps in six cities—Memphis, Tennessee; San Antonio, Texas; Bellevue, Washington; Tacoma, Washington; Seattle and Los Angeles.

The algorithm analyzed the imagery to determine whether the physical characteristics of neighborhoods (highways, crosswalks, different types of housing, parks, pet stores, greenways) correlated with reported obesity rates in different areas. The hypothesis, researchers said, was that amenities in different neighborhoods could affect a city’s propensity for obesity—meaning, for example, that residents who live in a place with high numbers of parks or gym facilities may be more likely to exercise regularly, improving the area’s overall health outcomes.

The data showed that neighborhoods with more space between buildings and higher numbers of parks and open spaces were likely to have lower rates of obesity. Those areas were typically wealthier, and researchers acknowledged that the algorithm may be skewed by both high and low income levels in a given area.

For example, the model underestimated the prevalence of obesity in Seattle, most likely due to the large number of green spaces in the region. Likewise, it overestimated the obesity rate in high-income parts of Memphis due to a lack of parks and green space, but residents there “can more readily afford gym memberships and other recreational facilities,” leading to a lower rate overall.

Still, researchers said, the algorithm is a first step toward predicting obesity rates using only visual imagery.

“Although our findings are likely to be explained at least in part by socioeconomic indicators, such as income, our analyses also suggest that the built environment features more consistently estimate obesity than per-capita income across all regions,” the report says. “The features of the built environment can be used in combination with other data sources for monitoring obesity prevalence, and these data could be useful for regions with delayed updates on obesity estimates and for programs focused on reducing obesity.”

Researchers hope to refine the algorithm to better account for income levels and other socioeconomic factors. The goal, according to the report, is to provide an inexpensive way to help city planners understand how construction can influence residents’ health.

“Neighborhood audits of features of the built environment have heretofore been conducted using costly and time-consuming on-site visits or neighborhood surveys,” the report says. “The development of data algorithms that can automatically process satellite images to create indicators of the built environment would dramatically lower the cost and allow for investigations of the effect of place characteristics on obesity prevalence.”

Kate Elizabeth Queram is a Staff Correspondent for Government Executive’s Route Fifty and is based in Washington, D.C.

NEXT STORY: Navigator Award Finalists: The Tech Innovators