Medical interns tend to have a higher rate of depression than the general population, due to factors such as high workload, irregular lifestyle, insufficient sleep, and lack of physical activity. We use mobile health data collected from medical interns and use it to forecast their mood, with the goal of tailoring interventions to improve mental health. Our data comes from the Intern Health Study, which includes data on mood, physical activity, sleep, heart rate, PHQ-9, geo-location, and baseline information. This time series data is collected from about 500 medical interns from multiple institutions throughout their entire internship period. Existing studies on mood prediction in mobile health mostly classify mood on a discrete scale, such as high mood vs. low mood. In contrast, we predict mood on a more continuous 1-10 scale. We use both linear methods and nonlinear methods and compare the performance of these methods. We see some signs that non-linear methods may perform better than linear methods. We also discuss the issues that arise to high rates of missingness in our data.
(This is joint work with a large team of collaborators who will be acknowledged during the talk)