Failing students saved by stress-detecting app

EARLIER this year, two students at Dartmouth College in New Hampshire who were due to be failed for skipping lectures and not completing assignments were spared the academic axe.

Why the leniency? According to an automated analysis of their smartphone data, both had stress and health-related issues they hadn't told their professors about. So instead of F's and a term's suspension, they were given a chance to complete the coursework over the summer and have now returned to campus.

The students have Andrew Campbell, a computer scientist at Dartmouth, and his colleagues to thank. The students, and 46 others, were enrolled in an experiment to see if data gathered from their phones could be used to guess their state of mind.

Campbell's team set out to discover why, out of a group of students arriving at university with similar qualifications, some excel while others miss lots of classes or even drop out entirely.

The researchers suspected that factors like the amount of sleep students get, their sociability, mood, workload and stress levels all played a role. So they built an app, called StudentLife, that monitors readings from smartphone sensors, and then recruited volunteers to use it over a 10-week term.

The app recorded almost every aspect of life that it was possible to measure, including physical activity levels, frequency and duration of conversations, and GPS location. The camera even watched for when the lights went out each night.

By crunching this data, the app could infer each student's levels of happiness, depression, loneliness and stress. That's possible because "flourishing" students, as the team calls them, are often with other people and have longer conversations, while depressed students interact less with others and have disrupted, or excessive, sleep. Loneliness is marked in part by mainly indoor activity, the team says, and the combination of disturbed sleep and short conversations is a predictor of stress.

The researchers compared these mental states with each student's performance, including grades for assignments and their grade-point average for the term.

"We found for the first time that passive and automatic sensor data, obtained from phones without any action by the user, significantly correlates student depression level, stress and loneliness with academic performance over the term," Campbell says. It also let them see how behaviour like gym usage and sleep times changed when students were faced with assignments or exams.

The results showed that students generally started the term in chipper moods, with most having lots of conversations, healthy sleep levels and busy activity patterns. As the term went on, workload increased, stress shot up – and sleep, chat and physical activity all dropped off. Daily interviews with volunteers confirmed that the automated analyses were accurate. Campbell will present the team's results at UbiComp in Seattle this week.

He believes the results are good evidence that phones will be able to provide continuous mental health assessment – much better than occasional questionnaires filled out when someone feeling depressed visits a doctor. And the app could work for people from all walks of life.

But accessing data on someone's every move will be controversial, even if it saves them their university place or job. "Privacy is the big issue here," says Cecilia Mascolo, who studies mobile sensing at the University of Cambridge. "You need to constrain this to a very specific application that will benefit people, and with the user always in control of their data." Still, she says, with proper protections in place, stress or depression could not only be detected, but also mitigated using information derived from phone sensors.

"People won't be given a prescription," she says. "They will be given an app.
New scientiest