BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/New_York
X-LIC-LOCATION:America/New_York
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260715T161118Z
LOCATION:Rhinelander Gallery
DTSTART;TZID=America/New_York:20260323T164500
DTEND;TZID=America/New_York:20260323T181500
UID:HFESHCS_2026 International Symposium on Human Factors and Ergonomics i
 n Health Care_sess106_POST210@linklings.com
SUMMARY:Decoding Mental Health Through Wearable Signals Using Machine Lear
 ning Approaches
DESCRIPTION:Guannan Liu, Jae Yoo, Gaojian Huang, Yuqi He, and Yue Luo (San
  Jose State University)\n\nIntroduction:\nMental health conditions are com
 mon and consequential: more than one in five U.S. adults experience a ment
 al illness annually, and worldwide the burden exceeds one billion people. 
 The impact is also substantial, including increased disability, higher mor
 tality, and major economic costs. In clinical and public health practice, 
 mental health is most often assessed through standardized screening and cl
 inical interviews. Binary risk stratification (e.g., screen-positive vs. s
 creen-negative) offers operational advantages for surveillance and triage.
  Instruments such as the PHQ-9 (≥10 for moderate depression) and GAD-7 (≥1
 0 for moderate anxiety) align with validated severity thresholds, guiding 
 care pathways and enabling consistent monitoring. Yet limitations remain: 
 self-report screeners are efficient and validated but episodic, subject to
  recall bias, and prone to false positives, while structured interviews, t
 hough diagnostic gold standards, are resource-intensive and time-consuming
 . These gaps have motivated interest in complementary, data-driven approac
 hes to mental health monitoring.\nRecent research emphasizes the value of 
 behavioral and physiological data streams for mental health assessment. Ph
 ysical activity and energy expenditure reveal movement patterns linked to 
 depression and anxiety risk; sleep duration and quality capture disturbanc
 es commonly associated with poor mental health; heart rate and other auton
 omic measures reflect stress reactivity, whereas sedentary behavior indica
 tes vulnerability. Importantly, these measures can be tracked continuously
  with wearable devices, opening opportunities for mental health assessment
  to extend beyond clinical settings into daily life.\nMore recently, artif
 icial intelligence (AI) techniques, particularly machine learning (ML), ha
 ve been applied to wearable data to reveal behavioral and physiological pa
 tterns that traditional statistical methods may overlook. For example, wea
 rable-based models for depression detection have reported accuracies appro
 aching 0.89, demonstrating promise. However, important challenges remain: 
 many studies rely on small, non-representative samples, device heterogenei
 ty complicates comparability, recruitment often skews toward younger and h
 ealthier populations, and most analyses remain cross-sectional, limiting i
 nsight into temporal dynamics. Addressing these issues will require larger
 , more diverse cohorts, standardized protocols, and rigorous external vali
 dation before wearable-based ML models can be reliably deployed in real-wo
 rld care.\nThis study aims to advance the field by leveraging the All of U
 s Research Program, a large-scale federally funded initiative linking elec
 tronic health records, survey responses, and wearable data from volunteer 
 participants across the U.S. Our work aims to classify mental health statu
 s (“positive vs. negative”) using wearable-derived features and to compare
  performance across canonical machine learning algorithms. In particular, 
 we evaluate the extent to which passively sensed physical activity and ene
 rgy expenditure can accurately distinguish mental health outcomes and disc
 uss their implications for human factors studies. \n\nApproach: \nThis stu
 dy leverages the database from the All of Us Research Program. Among the p
 articipants with the Fitbit dataset (15,620 of 838,000 participants, the l
 argest of its kind currently available), two data sources were analyzed: (
 i) self-rated mental health and (ii) daily physical activity features.\nMe
 ntal health was assessed with a five-point Likert item (poor–excellent), r
 ecorded into binary classes: negative (poor, fair) and positive (good, ver
 y good, excellent). The extracted cohort included 837 negative (11.5%) and
  6,455 positive (88.5%) cases. To address imbalance, all negative cases we
 re retained and positives were randomly under-sampled to a 1:1.5 ratio, yi
 elding a final dataset of 2,092 participants (837 negative, 1,255 positive
 ).\nPhysical activity and energy expenditure was represented by ten Fitbit
 -derived daily measures: (1) activity calories, (2) basal metabolic rate (
 BMR), (3) calories out, (4) elevation, (5) fairly active minutes, (6) ligh
 tly active minutes, (7) marginal calories, (8) sedentary minutes, (9) step
 s, and (10) very active minutes. Features were extracted for the seven day
 s preceding each survey, capturing temporal dynamics that may reflect fluc
 tuations in activity predictive of mental health. Age was included as a co
 variate, with results showing that participants with positive mental healt
 h were, on average, older than those with negative mental health.\nWe then
  trained five supervised machine learning models, namely logistic regressi
 on (LR), support vector machines (SVM), decision tree (DT), random forest 
 (RF), and extreme gradient boosting (XGBClassifier). Each participant was 
 represented by seven-day activity sequences (10 × 7 features) plus age. Mo
 dels were tuned using five-fold cross-validation and evaluated on a held-o
 ut test set using area under the ROC curve (AUC), accuracy, precision, rec
 all, and F1 score as the performance metric.\n\nFindings: \nThe results re
 vealed clear differences in model performance. LR, SVM, and RF achieved co
 mparable levels of discrimination, each reaching an AUC of approximately 0
 .75. These outcomes indicate that both linear methods (LR and SVM) and an 
 ensemble-based tree method (RF) were able to capture meaningful patterns i
 n the wearable-derived features for distinguishing between positive and ne
 gative mental health states. XGBClassifier, another ensemble-based approac
 h, performed slightly better, achieving an AUC of 0.77. Although the impro
 vement over other models was modest, it suggests that gradient boosting ca
 n exploit non-linear interactions and complex feature relationships more e
 ffectively. In contrast, the DT model lagged behind, with an AUC of 0.63, 
 reflecting weaker discriminative ability and confirming the known tendency
  of single-tree classifiers to overfit and perform poorly on unseen data.\
 nWhen evaluating other performance metrics, XGBClassifier again achieved t
 he highest overall performance. It achieved the highest F1 score (0.77) an
 d recall (0.88). This indicates effectiveness in correctly identifying par
 ticipants with good mental health while maintaining a balance between sens
 itivity and precision, thus keeping the false positive rate manageable. Su
 ch reliability is important in public health and clinical contexts, where 
 accurate stratification supports appropriate allocation of resources and f
 ollow-up care. RF also achieved comparable results (F1 score = 0.77 and re
 call = 0.84) and recorded the highest accuracy (0.71) and precision (0.71)
 . By balancing precision, recall, and F1, it provided a robust alternative
  to XGBClassifier among the ensemble models. LR and SVM showed nearly iden
 tical performance, demonstrating the robustness of linear models, whereas 
 DT showed the weakest accuracy and F1 score, indicating limited generaliza
 bility.\nTaken together, these findings suggest that ensemble-based method
 s (RF and XGBClassifier) offer distinct advantages over simpler classifier
 s by leveraging multiple decision rules and capturing complex, non-linear 
 patterns in the data. \n\nTakeaways From Human Factors Perspectives:\nWear
 able data combined with ML models offer promising applications in human fa
 ctors research by enabling continuous, unobtrusive monitoring of mental he
 alth in naturalistic settings. Such assessments can support the design of 
 adaptive systems that detect early signs of mental strain and guide interv
 entions in safety-critical domains such as healthcare, transportation, and
  aviation. In healthcare, daily-life monitoring of activity and energy exp
 enditure may aid in identifying poor mental health symptoms, tracking reco
 very, and supporting adherence to treatment plans. In transportation, reco
 gnizing positive or negative mental health states can help flag risks such
  as stress, anxiety, or other emotional regulation that impair attention a
 nd decision-making, enabling timely interventions to improve road safety.\
 nFrom a human factors perspective, these findings underscore both opportun
 ities and challenges in integrating wearable-based mental health assessmen
 t into practice. Ensemble models such as XGBClassifier and RF demonstrate 
 strong predictive performance, yet their limited interpretability remains 
 a critical consideration for human-centered system design. Transparency an
 d user trust are essential if such models are to meaningfully inform decis
 ion-making or guide interventions. Beyond detecting positive mental health
  states, the capacity to capture true negatives—accurately identifying poo
 r mental health—is equally important to ensure that at-risk individuals ar
 e not overlooked. Looking forward, future work could also incorporate addi
 tional features such as sleep quality and sedentary behavior, which may ca
 pture richer behavioral dynamics, as well as temporal dependencies across 
 weekly cycles or circadian rhythms to model fluctuations more effectively.
  To advance human-centered systems, further considerations include enhanci
 ng interpretability for both clinicians and end users, ensuring representa
 tiveness across populations, safeguarding privacy, and designing feedback 
 mechanisms that are supportive rather than stigmatizing.\n\nTrack: Digital
  Health, Hospital Environments, Patient Safety Research and Initiatives, M
 edical and Drug Delivery Devices, Simulation and Education\n\n
END:VEVENT
END:VCALENDAR