The interior of the Fitbit office at the Boston waterfront has the feel of a Silicon Valley start-up -- open workspaces, a ping-pong table, multiple couches, and a fully-stocked snack bar with free chips, candy, roasted almonds, single-serve juices, and a keg. When a team of CDS (Center for Data Science) students and staff arrived in early May, sunlight streamed in from tall windows on three sides of the open office space.
Brant Cheikes, Executive Director of the CDS, was there to give a presentation to Fitbit staff, along with CICS (College of Information and Computer Sciences) Master’s students Ravi Agrawal, Abhinav Shaw, and Rasmus Christiansen. Also in attendance was Jennifer Page, CDS Associate Director of Strategic Communications. (Master’s student Rahul Handa worked on the project but was unable to attend.) The focus of the presentation was a research study that the students are working on this summer as part of the CDS Data Science for Common Good program. The program matches Master’s students with a real-world “common good” problem to be worked on during the summer session. In this case, the motivating problem is the creation a predictive model of student stress using wearable devices like Fitbits, to improve student outcomes like academic performance, retention and graduation rates.
Cheikes first gave an overview of CDS, covering focus areas, major research support, and the Industry Affiliates Program. Then the students took over with their presentation on the StudentLife paper published by Dartmouth College1, and how it can be enhanced using data science and predictive modeling. The StudentLife project was a small-scale study that used a smartphone app to track the level of stress amongst 48 Dartmouth students over a 10-week period. The CDS team applied four models to the StudentLife dataset, and established that long short-term memory (LSTM) neural networks provide the best predictive model of student stress. The team then laid out their plans to improve the model, and design a new research study this summer, using Fitbit Ionic devices to collect data. After the presentation the CDS team met via teleconference with Fitbit data scientists in their San Francisco headquarters. The Fitbit team is very interested in the study and is looking forward to future collaboration with CDS.