Study predicts obesity levels in cities based on infrastructure

Dive Brief:

Researchers at the University of Washington have created an artificial intelligence (AI) algorithm that estimates citizens' obesity levels within a city based on the city's infrastructure, indicates their study published in JAMA Network Open.
The study used AI on almost 150,000 Google Maps satellite and street view images from six cities: Bellevue, WA; Los Angeles; Memphis, TN; San Antonio, TX; Seattle; and Tacoma, WA.
The study looked at physical environment characteristics like types of housing, building spacing, crosswalks, highways and the presence of green space, and how those characteristics may link to weight.

Dive Insight:

A number of scientific studies have made significant discoveries and conclusions in recent years by using various algorithms on Google Map images. Using AI takes away potential human bias and leads to a more consistent analysis.

This study could lead to further analyses that would help city planners identify infrastructure shortcomings that could be improved. Doing the analysis with AI proves more efficient and cost effective than relying on humans to examine and document all of the physical environment factors, in addition to providing consistency. Beyond improving infrastructure, the analysis could identify specific neighborhoods that could benefit from additional municipal health, fitness and recreational programs.

The University of Washington researchers touted significant potential for their study, but they also pointed out some shortcomings. The AI might have weighted certain factors too heavily in analyzing obesity rates or might not have considered real-life conditions. Seattle's obesity rate, for example, was underestimated in part because of the large amount of green space within the city.

The algorithm also can be skewed by income levels, both high and low. The researchers noted that some high-income neighborhoods, such as in Memphis, were viewed as having high obesity rates because their infrastructure does not appear to promote physical activity. However people living in those neighborhoods are more able to afford gym memberships or paying for other recreational activities that keep them active.

When considering the areas for improvement, the researchers stated that this AI model can be used in combination with other data sources to monitor an area's obesity prevalence. They expressed the desire to further refine the system to create more consistency and account for socioeconomic status.