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.
more »
« less
Machine learning for passive mental health symptom prediction: Generalization across different longitudinal mobile sensing studies
Mobile sensing data processed using machine learning models can passively and remotely assess mental health symptoms from the context of patients’ lives. Prior work has trained models using data from single longitudinal studies, collected from demographically homogeneous populations, over short time periods, using a single data collection platform or mobile application. The generalizability of model performance across studies has not been assessed. This study presents a first analysis to understand if models trained using combined longitudinal study data to predict mental health symptoms generalize across current publicly available data. We combined data from the CrossCheck (individuals living with schizophrenia) and StudentLife (university students) studies. In addition to assessing generalizability, we explored if personalizing models to align mobile sensing data, and oversampling less-represented severe symptoms, improved model performance. Leave-one-subject-out cross-validation (LOSO-CV) results were reported. Two symptoms (sleep quality and stress) had similar question-response structures across studies and were used as outcomes to explore cross-dataset prediction. Models trained with combined data were more likely to be predictive (significant improvement over predicting training data mean) than models trained with single-study data. Expected model performance improved if the distance between training and validation feature distributions decreased using combined versus single-study data. Personalization aligned each LOSO-CV participant with training data, but only improved predicting CrossCheck stress. Oversampling significantly improved severe symptom classification sensitivity and positive predictive value, but decreased model specificity. Taken together, these results show that machine learning models trained on combined longitudinal study data may generalize across heterogeneous datasets. We encourage researchers to disseminate collected de-identified mobile sensing and mental health symptom data, and further standardize data types collected across studies to enable better assessment of model generalizability.
more »
« less
- Award ID(s):
- 1750326
- NSF-PAR ID:
- 10325547
- Editor(s):
- Chen, Chi-Hua
- Date Published:
- Journal Name:
- PLOS ONE
- Volume:
- 17
- Issue:
- 4
- ISSN:
- 1932-6203
- Page Range / eLocation ID:
- e0266516
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
null (Ed.)The prevalence of mobile phones and wearable devices enables the passive capturing and modeling of human behavior at an unprecedented resolution and scale. Past research has demonstrated the capability of mobile sensing to model aspects of physical health, mental health, education, and work performance, etc. However, most of the algorithms and models proposed in previous work follow a one-size-fits-all (i.e., population modeling) approach that looks for common behaviors amongst all users, disregarding the fact that individuals can behave very differently, resulting in reduced model performance. Further, black-box models are often used that do not allow for interpretability and human behavior understanding. We present a new method to address the problems of personalized behavior classification and interpretability, and apply it to depression detection among college students. Inspired by the idea of collaborative-filtering, our method is a type of memory-based learning algorithm. It leverages the relevance of mobile-sensed behavior features among individuals to calculate personalized relevance weights, which are used to impute missing data and select features according to a specific modeling goal (e.g., whether the student has depressive symptoms) in different time epochs, i.e., times of the day and days of the week. It then compiles features from epochs using majority voting to obtain the final prediction. We apply our algorithm on a depression detection dataset collected from first-year college students with low data-missing rates and show that our method outperforms the state-of-the-art machine learning model by 5.1% in accuracy and 5.5% in F1 score. We further verify the pipeline-level generalizability of our approach by achieving similar results on a second dataset, with an average improvement of 3.4% across performance metrics. Beyond achieving better classification performance, our novel approach is further able to generate personalized interpretations of the models for each individual. These interpretations are supported by existing depression-related literature and can potentially inspire automated and personalized depression intervention design in the futuremore » « less
-
more » « less
Background Maternal loneliness is associated with adverse physical and mental health outcomes for both the mother and her child. Detecting maternal loneliness noninvasively through wearable devices and passive sensing provides opportunities to prevent or reduce the impact of loneliness on the health and well-being of the mother and her child.
Objective The aim of this study is to use objective health data collected passively by a wearable device to predict maternal (social) loneliness during pregnancy and the postpartum period and identify the important objective physiological parameters in loneliness detection.
Methods We conducted a longitudinal study using smartwatches to continuously collect physiological data from 31 women during pregnancy and the postpartum period. The participants completed the University of California, Los Angeles (UCLA) loneliness questionnaire in gestational week 36 and again at 12 weeks post partum. Responses to this questionnaire and background information of the participants were collected through our customized cross-platform mobile app. We leveraged participants’ smartwatch data from the 7 days before and the day of their completion of the UCLA questionnaire for loneliness prediction. We categorized the loneliness scores from the UCLA questionnaire as loneliness (scores≥12) and nonloneliness (scores<12). We developed decision tree and gradient-boosting models to predict loneliness. We evaluated the models by using leave-one-participant-out cross-validation. Moreover, we discussed the importance of extracted health parameters in our models for loneliness prediction.
Results The gradient boosting and decision tree models predicted maternal social loneliness with weighted F1-scores of 0.897 and 0.872, respectively. Our results also show that loneliness is highly associated with activity intensity and activity distribution during the day. In addition, resting heart rate (HR) and resting HR variability (HRV) were correlated with loneliness.
Conclusions Our results show the potential benefit and feasibility of using passive sensing with a smartwatch to predict maternal loneliness. Our developed machine learning models achieved a high F1-score for loneliness prediction. We also show that intensity of activity, activity pattern, and resting HR and HRV are good predictors of loneliness. These results indicate the intervention opportunities made available by wearable devices and predictive models to improve maternal well-being through early detection of loneliness.
-
more » « less
Abstract This study investigates the generalizability and predictive validity of associations between gastrointestinal (GI) symptoms and youth anxiety to establish their utility in community mental health decision‐making. We analyzed data from youth ages 3 to 21 years in volunteer cohorts collected in Los Angeles (N = 327) and New York City (N = 102), as well as the Healthy Brain Network cohort (N = 1957). Youth GI distress was measured through items taken from the parent‐reported Child Behavior Checklist (CBCL). We examined generalizability of GI–anxiety associations across cohorts and anxiety reporters, then evaluated the performance of these models in predicting youth anxiety in holdout data. Consistent with previous work, higher levels of gastrointestinal distress were associated with more parent‐reported youth anxiety behaviors in all three cohorts. Models trained on data from the Healthy Brain Network cohort predicted parent‐reported and child‐reported anxiety behaviors, as well as clinician‐evaluated anxiety diagnoses, at above chance levels in holdout data. Models which included GI symptoms often, but not always, outperformed models based on age and sex alone in predicting youth anxiety. Based on the generalizability and predictive validity of GI–anxiety associations investigated here, GI symptoms may be an effective tool for child‐facing professionals for identifying children at risk for anxiety (Preprint:
https://psyarxiv.com/zgavu/ ). -
Abstract. Advances in ambient environmental monitoring technologies are enabling concerned communities and citizens to collect data to better understand their local environment and potential exposures. These mobile, low-cost tools make it possible to collect data with increased temporal and spatial resolution, providing data on a large scale with unprecedented levels of detail. This type of data has the potential to empower people to make personal decisions about their exposure and support the development of local strategies for reducing pollution and improving health outcomes. However, calibration of these low-cost instruments has been a challenge. Often, a sensor package is calibrated via field calibration. This involves colocating the sensor package with a high-quality reference instrument for an extended period and then applying machine learning or other model fitting technique such as multiple linear regression to develop a calibration model for converting raw sensor signals to pollutant concentrations. Although this method helps to correct for the effects of ambient conditions (e.g., temperature) and cross sensitivities with nontarget pollutants, there is a growing body of evidence that calibration models can overfit to a given location or set of environmental conditions on account of the incidental correlation between pollutant levels and environmental conditions, including diurnal cycles. As a result, a sensor package trained at a field site may provide less reliable data when moved, or transferred, to a different location. This is a potential concern for applications seeking to perform monitoring away from regulatory monitoring sites, such as personal mobile monitoring or high-resolution monitoring of a neighborhood. We performed experiments confirming that transferability is indeed a problem and show that it can be improved by collecting data from multiple regulatory sites and building a calibration model that leverages data from a more diverse data set. We deployed three sensor packages to each of three sites with reference monitors (nine packages total) and then rotated the sensor packages through the sites over time. Two sites were in San Diego, CA, with a third outside of Bakersfield, CA, offering varying environmental conditions, general air quality composition, and pollutant concentrations. When compared to prior single-site calibration, the multisite approach exhibits better model transferability for a range of modeling approaches. Our experiments also reveal that random forest is especially prone to overfitting and confirm prior results that transfer is a significant source of both bias and standard error. Linear regression, on the other hand, although it exhibits relatively high error, does not degrade much in transfer. Bias dominated in our experiments, suggesting that transferability might be easily increased by detecting and correcting for bias. Also, given that many monitoring applications involve the deployment of many sensor packages based on the same sensing technology, there is an opportunity to leverage the availability of multiple sensors at multiple sites during calibration to lower the cost of training and better tolerate transfer. We contribute a new neural network architecture model termed split-NN that splits the model into two stages, in which the first stage corrects for sensor-to-sensor variation and the second stage uses the combined data of all the sensors to build a model for a single sensor package. The split-NN modeling approach outperforms multiple linear regression, traditional two- and four-layer neural networks, and random forest models. Depending on the training configuration, compared to random forest the split-NN method reduced error 0 %–11 % for NO2 and 6 %–13 % for O3.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1371/journal.pone.0266516
