Five Million App Users Show How the World Kept Moving During COVID

Scientists studied exercise information from more than five million people. These people lived in sixty-five countries. They used a smartphone app that tracks physical activity. The data spanned from January 2019 to June 2021. This period included the beginning of the COVID-19 pandemic.

The goal was to see who kept exercising. Health guidelines say adults need at least one hundred fifty minutes of movement each week. Researchers built computer programs to guess if a person would meet that target. They tried several types of artificial intelligence. Some programs looked only at past workouts. Others added details like age, local weather, and pandemic rules.

One test happened in Middlesex County, Massachusetts. There, a graph neural network performed best. It predicted correctly about eighty-nine point nine percent of the time before the pandemic. During the pandemic, its accuracy dropped to about eighty-two point one percent. Across the whole world, a standard neural network was right about eighty-seven point six percent before, and eighty-two point eight percent during.

The strongest clue was a person's own exercise history. Knowing what someone did in the previous one to three weeks improved accuracy by twenty percentage points. Adding social network data or making the models more complex helped only a little.

This research shows that app data can track health habits quickly. When the next major health crisis arrives, officials could use similar information. They might see where people are struggling to stay active. Then they could create better ways to help communities keep moving.