SmartGPA is a cloud-backed app that relies on embedded passive sensors as well as special algorithms that can determine behaviors by the phone user, from studying to partying to face-to-face-communications to sleep. That information can then be crunched to predict students' GPA within 17 hundreds of a point, according to Andrew T. Campbell, who co-authored paper on the research with colleagues from Dartmouth and the University of Texas at Austin.
MORE:10 Cool Networking & Computing Research Projects
Their paper, "SmartGPA: How Smartphones Can Assess and Predict Academic Performance of College Students," will be presented at the ACM UbiComp conference on pervasive and ubiquitous computing in September in Japan. The study was based on 30 undergrads using the app for 10 weeks at Dartmouth, and an expanded study is planned for both Dartmouth and in Texas. (A video below further explains the research, in a non-technical way.)
Campbell, whose pedestrian safety app I previously wrote about, started work on SmartGPA because of this perhaps troubling discovery: "I previously found no correlation between class attendance and GPA." More broadly, the study aims for insight into "why students with similar academic capability at the same institution do better or worse than one another."
SmartGPA runs entirely in the background of a smartphone, using embedded sensor signals and machine learning technology to infer user behavior, without users having to take any special action, Campbell says. Studying could be inferred, for example, if the student were using his or her phone in an area determined to be a study spot, and also based on the ambient sound in such an area. Usage of other apps, such as social media ones, is recorded, but SmartGPA doesn't dive into the actual interactions within such apps.
Why is the SmartGPA app only Android based It turns out that you can't use continuous background sensing on the iPhone without jailbreaking it. "Apple needs to get with the program: They'll lose out to other advances in the new wave of predictive apps like ours," Campbell warns.
The new study comes to many rather unsurprising conclusions about what leads to better school performance: those who partied less and studied more as the term progressed, for example, did better. But other less apparent findings were also revealed: those who had fewer face-to-face conversations at night did better, as did those whose stress levels rose toward mid-term and fell toward end-of-term.
The breakthrough with SmartGPA is that the team has developed "a core mobile sensing technology to determine user behavior continuously with no user interaction," Campbell says.
"We've applied it to 'performance' of students -- GPA. But this core could be used to determine the performance of a software engineer, Uber driver or teacher. That's the idea. Coming up with different models for different types of careers and jobs," he says.
Expect an app to become generally available a year or so from now, once the current technology has been tested at more places and at scale "in more diverse places than the Ivies." Not only will the app be able to predict performance, but will suggest changes in behavior to improve upon it.