Search for: All records

Longitudinal human behavior modeling has received increasing attention over the years due to its widespread applications to patient monitoring, dietary and lifestyle recommendations, and just-in-time intervention for at-risk individuals (e.g., prob- lematic drug users and struggling students), to name a few. Using in-the-moment health data collected via ubiquitous devices (e.g., smartphones and smartwatches), this multidisciplinary field focuses on developing predictive models for certain health or well-being outcomes (e.g., depression and stress) in the short future given the time series of individual behaviors (e.g., resting heart rate, sleep quality, and current feelings). Yet, most existing models on these data, which we refer to as ubiquitous health data, do not achieve adequate accuracy. The latest works that yielded promising results have yet to consider realistic aspects of ubiquitous health data (e.g., containing features of different types and high rate of missing values) and the consumption of various resources (e.g., computing power, time, and cost). Given these two shortcomings, it is dubious whether these studies could translate to realistic settings. In this paper, we propose MuHBoost, a multi-label boosting method for addressing these shortcomings, by leveraging advanced methods in large language model (LLM) prompting and multi-label classification (MLC) to jointly predict multiple health or well-being outcomes. Because LLMs can hal- lucinate when tasked with answering multiple questions simultaneously, we also develop two variants of MuHBoost that alleviate this issue and thereby enhance its predictive performance. We conduct extensive experiments to evaluate MuH- Boost and its variants on 13 health and well-being prediction tasks defined from four realistic ubiquitous health datasets. Our results show that our three developed methods outperform all considered baselines across three standard MLC metrics, demonstrating their effectiveness while ensuring resource efficiency. 

BackgroundSubstance use induces large economic and societal costs in the U.S. Understanding the change in substance use behaviors of persons who use drugs (PWUDs) over time, therefore, is important in order to inform healthcare providers, policymakers, and other stakeholders toward more efficient allocation of limited resources to at-risk PWUDs. ObjectiveThis study examines the short-term (within a year) behavioral changes in substance use of PWUDs at the population and individual levels. Methods237 PWUDs in the Great Plains of the U.S. were recruited by our team. The sample provides us longitudinal survey data regarding their individual attributes, including drug use behaviors, at two separate time periods spanning 4-12 months. At the population level, we analyze our data quantitatively for 18 illicit drugs; then, at the individual level, we build interpretable machine learning logistic regression and decision tree models for identifying relevant attributes to predict, for a given PWUD, (i) which drug(s) they would likely use and (ii) which drug(s) they would likely increase usage within the next 12 months. All predictive models were evaluated by computing the (averaged) Area under the Receiver Operating Characteristic curve (AUROC) and Area under the Precision-Recall curve (AUPR) on multiple distinct sets of hold-out sample. ResultsAt the population level, the extent of usage change and the number of drugs exhibiting usage changes follow power-law distributions. At the individual level, AUROC’s of the models for the top-4 prevalent drugs (marijuana, methamphetamines, amphetamines, and cocaine) range 0.756-0.829 (+2.88-7.66% improvement with respect to baseline models using only current usage of the respective drugs as input) for (i) and 0.670-0.765 (+4.34-18.0%) for (ii). The corresponding AUPR’s of the said models range 0.729-0.947 (+2.49-13.6%) for (i) and 0.348-0.618 (+26.9-87.6%) for (ii). ConclusionThe observed qualitative changes in short-term substance usage and the trained predictive models for (i) and (ii) can potentially inform human decision-making toward efficient allocation of appropriate resources to PWUDs at highest risk. 

Substance use is a global issue that negatively impacts millions of persons who use drugs (PWUDs). In practice, identifying vulnerable PWUDs for efficient allocation of appropriate resources is challenging due to their complex use patterns (e.g., their tendency to change usage within months) and the high acquisition costs for collecting PWUD-focused substance use data. Thus, there has been a paucity of machine learning models for accurately predicting short-term substance use behaviors of PWUDs. In this paper, using longitudinal survey data of 258 PWUDs in the U.S. Great Plains collected by our team, we design a novel GAN that deals with high-dimensional low-sample-size tabular data and survey skip logic to augment existing data to improve classification models' prediction on (A) whether the PWUDs would increase usage and (B) at which ordinal frequency they would use a particular drug within the next 12 months. Our evaluation results show that, when trained on augmented data from our proposed GAN, the classification models improve their predictive performance (AUROC) by up to 13.4% in Problem (A) and 15.8% in Problem (B) for usage of marijuana, meth, amphetamines, and cocaine, which outperform state-of-the-art generative models. 

In this paper, we present a framework for quantifying the impact of interventions on the full trajectories of students' experiences. The interventions are given periodically based on student performance forecasting from an artificial intelligence (AI) model. We performed a small-scale randomized controlled trial for evaluating the impact of the AI-based intervention system on the undergraduate students of a science, technology, engineering, and mathematics (STEM) course. Intervention messaging content was based on machine learning forecasting models trained on data collected from the students in the same course over the preceding 3 years. Trial results show that the intervention produced a statistically significant increase in the proportion of students that achieved a passing grade. By applying the trajectory-analysis framework we find that the intervention impacts the stories of some types of students more than others, and use this to define new ways of identifying students who are most likely to benefit. Together these outcomes point to the potential and promise of just-in-time interventions for STEM learning and the need for larger fully-powered randomized controlled trials. 

We present results from a small-scale randomized controlled trial that evaluates the impact of just-in-time interventions on the academic outcomes of N = 65 undergraduate students in a STEM course. Intervention messaging content was based on machine learning forecasting models of data collected from 537 students in the same course over the preceding 3 years. Trial results show that the intervention produced a statistically significant increase in the proportion of students that achieved a passing grade. The outcomes point to the potential and promise of just-in-time interventions for STEM learning and the need for larger fully-powered randomized controlled trials.