GLOBEM: Cross-Dataset Generalization of Longitudinal Human Behavior Modeling

Xuhai Xu, Xin Liu, Han Zhang, Weichen Wang, Subigya Nepal, Yasaman S. Sefidgar, Woosuk Seo, Kevin S. Kuehn, Jeremy F. Huckins, Margaret E. Morris, Paula S. Nurius, Eve A. Riskin, Shwetak N. Patel, Tim Althoff, Andrew Campbell, Anind K. Dey, and Jennifer Mankoff. GlOBEM: Cross-Dataset Generalization of Longitudinal Human Behavior Modeling. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 6(4): 190:1-190:34 (2022).

There is a growing body of research revealing that longitudinal passive sensing data from smartphones and wearable devices can capture daily behavior signals for human behavior modeling, such as depression detection. Most prior studies build and evaluate machine learning models using data collected from a single population. However, to ensure that a behavior model can work for a larger group of users, its generalizability needs to be verified on multiple datasets from different populations. We present the first work evaluating cross-dataset generalizability of longitudinal behavior models, using depression detection as an application. We collect multiple longitudinal passive mobile sensing datasets with over 500 users from two institutes over a two-year span, leading to four institute-year datasets. Using the datasets, we closely re-implement and evaluated nine prior depression detection algorithms. Our experiment reveals the lack of model generalizability of these methods. We also implement eight recently popular domain generalization algorithms from the machine learning community. Our results indicate that these methods also do not generalize well on our datasets, with barely any advantage over the naive baseline of guessing the majority. We then present two new algorithms with better generalizability. Our new algorithm, Reorder, significantly and consistently outperforms existing methods on most cross-dataset generalization setups. However, the overall advantage is incremental and still has great room for improvement. Our analysis reveals that the individual differences (both within and between populations) may play the most important role in the cross-dataset generalization challenge. Finally, we provide an open-source benchmark platform GLOBEM – short for Generalization of LOngitudinal BEhavior Modeling – to consolidate all 19 algorithms. GLOBEM can support researchers in using, developing, and evaluating different longitudinal behavior modeling methods. We call for researchers’ attention to model generalizability evaluation for future longitudinal human behavior modeling studies.

Distress and resilience among marginalized undergraduates

Nurius, P. S., Sefidgar, Y. S., Kuehn, K. S, Jake, X, Zhang, H., Browning, A., Riskin, E., Dey, A. K., & Mankoff, J.  Distress among undergraduates: Marginality, stressors and resilience supports. Journal of American College Health, 1-9.

Stress and related mental health struggles are of growing concern at colleges and universities across the country and internationally, with some evidence of levels higher than general population peers. The university experience can pose considerable strain on students, in some cases adding to early and current life stressors, and, if not mitigated, can lead to impaired well-being and academic success/retention.

This study provides a 2019 data snapshot of multiple stressor effects on early-stage students, resilience resources (or the lack thereof) that can mitigate these effects, and sociodemographic characteristics reflecting minoritized identities. Participants were 253 first- and second-year undergraduate students (age =18.76; 49.80% male, 69% students of color) enrolled at the University of Washington.

Multivariate analysis demonstrated significant associations between greater stress exposures and lower levels of resilience resources with each of three mental health indicators—perceived stress (intensity of experienced stress), depression, and anxiety. Stressors such as poor physical health, discrimination exposure, experiencing one or more marginalizing status (e.g., first generation student, having disabilities, sexual minority), and using maladaptive coping strategies (e.g, denial, self-blame) significantly accounted for each of the mental health indicators. Prior stressors such as adverse childhood experiences and other life and academic adversities were also significantly correlated with the mental health variables.

Race/ethnicity was less clearly patterned, although students of Asian descent reported significantly greater depression and anxiety, and females reported higher levels on all distress forms. In terms of resilience supports, those reporting greater social support and perception of oneself as a “bounce back” kind of person reported lesser psychological distress and these variables reduced the effects of stressors. Assessment of student well-being from this same project during the 2020 COVID-19 context indicated that students entering the pandemic with mental health vulnerabilities experienced significantly greater psychological distress and academic strain as the university pivoted toward remote instruction, signaling highly consequential differences (Morris et al., 2021)

These results support the value of “poly-strengths” –multiple forms of resilience- fostering resources–for mitigating the effects of stressors on psychological distress. College leaders are noting increases in the severity of students’ mental health concerns and demand for services, changing the roles of campus counseling centers, and requiring new institutional responses. Better understanding cumulative stress/resilience resource profiles, particularly among marginalized students and those experiencing discrimination, can help universities in prioritizing institutional support responses toward prevention, strengthening resilience, and mitigating psychological distress.