About

Nigel Bosch is an associate professor in the School of Information Sciences and the Department of Educational Psychology at the University of Illinois Urbana-Champaign. He is also a faculty affiliate at the National Center for Supercomputing Applications (NCSA) and Illinois Informatics. His research focuses on machine learning and data mining methods to study human behaviors, particularly in learning contexts, with an emphasis on learning analytics, model generalization, and fairness and transparency in machine learning. Bosch's work involves analyzing data such as facial expressions, audio recordings, log files of user actions, and other sources that provide insights into learners' behaviors. He aims to develop fairer learning software and research methods by analyzing biases in these machine learning techniques.

Research topics

• Machine Learning
• Computer Science
• Artificial Intelligence
• Psychology
• Natural Language Processing
• Applied psychology
• Social psychology
• Simulation
• Data science
• Biology
• Clinical psychology
• Developmental psychology
• Pharmacology
• Medicine

Selected publications

• A Framework for Considering Exploration, Interpretation, and Confirmation During Data Analysis: Computationally Assisted Analysis of Teacher-Group Interactions

  Zenodo (CERN European Organization for Nuclear Research) · 2026-03-03

  articleOpen accessSenior author

  Education researchers increasingly analyze heterogeneous, multimodal data with computational tools. Yet, reporting rarely makes explicit who (human or computer) leads meaning-making at different points in the analysis. We introduce a framework for analytic agency that distinguishes three stages, exploration, interpretation, and confirmation, and classifies each as primarily human- or computer-led, as considering stage-level leadership can clarify assumptions in analysis. We demonstrate the framework in a multimodal case study of teacher-student group interactions in high school mathematics classrooms. Using 15 classroom videos from three teachers, we selected 21 student groups and developed a pose-based detector that flags interactions. The pipeline aligned group-level audio and word-level transcripts to each detected window and computed acoustic/prosodic features and large-language-model indicators for question-asking, confusion, help-seeking, and math talk. Across the corpus, the detector surfaced 317 interaction events (M = 15.10 per group, SD = 12.42; mean duration = 32.73s). We compared before, during, and after segments using paired tests and mixed-effects models. Naturally, results for mixed-effects models showed significant shifts in keypoints before-to-during and before-to-after for those emphasized in the detection approach, while audio features showed no significant changes. One transcript indicator, confusion, decreased after interactions (beta = -0.061, p = .049). The pipeline showed preferences for spatial co-presence rather than interaction discourse change, which illustrates how leadership in exploration shaped what became detectable and, consequently, how interpretation proceeded. In the paper's conclusion, we outline hybrid, iterative variants and discuss limitations. Making stage-level agency explicit can help researchers align methodological choices with theoretical aims and produce more transparent, auditable analyses of complex classroom data.

  PublisherOA PDFDOI

• A counterfactual explainable AI-driven approach to support students’ self-regulated learning in computer-based learning environments

  Open MIND · 2026-01-01

  otherOpen access

  In this study, we propose a counterfactual explainable AI-driven self-regulated learning (SRL) training system for computer-based learning environments. The system uses real-time log data to generate an AI-predicted final test score and provides a personalized, interactive study guide that helps students reflect on their current SRL engagement and plan how to allocate time across SRL-relevant strategies. Students can set a desired final test score and explore "what-if" scenarios through counterfactual recommendations (e.g., specific changes to their learning strategies needed to reach the desired score) generated using Diverse Counterfactual Explanations (DiCE). We will evaluate the system in a between-subjects randomized controlled trial with 190 participants. We will examine whether receiving the counterfactual-driven SRL training improves learning gains and metacognitive calibration, and increases engagement in metacognitive strategies.

  PublisherOA PDFDOI

• A Framework for Considering Exploration, Interpretation, and Confirmation During Data Analysis: Computationally Assisted Analysis of Teacher-Group Interactions

  Zenodo (CERN European Organization for Nuclear Research) · 2026-03-03

  articleOpen accessSenior author

  Education researchers increasingly analyze heterogeneous, multimodal data with computational tools. Yet, reporting rarely makes explicit who (human or computer) leads meaning-making at different points in the analysis. We introduce a framework for analytic agency that distinguishes three stages, exploration, interpretation, and confirmation, and classifies each as primarily human- or computer-led, as considering stage-level leadership can clarify assumptions in analysis. We demonstrate the framework in a multimodal case study of teacher-student group interactions in high school mathematics classrooms. Using 15 classroom videos from three teachers, we selected 21 student groups and developed a pose-based detector that flags interactions. The pipeline aligned group-level audio and word-level transcripts to each detected window and computed acoustic/prosodic features and large-language-model indicators for question-asking, confusion, help-seeking, and math talk. Across the corpus, the detector surfaced 317 interaction events (M = 15.10 per group, SD = 12.42; mean duration = 32.73s). We compared before, during, and after segments using paired tests and mixed-effects models. Naturally, results for mixed-effects models showed significant shifts in keypoints before-to-during and before-to-after for those emphasized in the detection approach, while audio features showed no significant changes. One transcript indicator, confusion, decreased after interactions (beta = -0.061, p = .049). The pipeline showed preferences for spatial co-presence rather than interaction discourse change, which illustrates how leadership in exploration shaped what became detectable and, consequently, how interpretation proceeded. In the paper's conclusion, we outline hybrid, iterative variants and discuss limitations. Making stage-level agency explicit can help researchers align methodological choices with theoretical aims and produce more transparent, auditable analyses of complex classroom data.

  PublisherDOI

• Prompting for Teachability: Designing Novice Personas in LLMs for Learning by Teaching Contexts

  2026-04-25

  articleOpen accessSenior author

  Learning by teaching (LbT) is a well-established instructional framework in which students deepen understanding by explaining material to a peer or tutee. Large Language Models (LLMs) create new opportunities to scale LbT by simulating novice learners, but their default tendency toward expert-like responses risks undermining the tutor's role. This study investigates which prompting strategies most effectively elicit novice-behavior from LLMs in writing-related domains. We generated 30,720 combined prompts across five domains and evaluated three models (Qwen3-235B, Llama 4, Kimi-K2) using both multiple-choice quizzes and short persuasive essays. Outputs were scored on quiz accuracy, essay quality, and essay persuasiveness using an AI-judge rubric. Regression analysis revealed a clear pattern: constraint prompts that explicitly forced error production consistently outperformed persona-, misconception‑, and uncertainty-based prompts. Across both quiz and essay outcomes, direct commands to “answer incorrectly” or “get 2–3 wrong” yielded the strongest novice-like behavior, while indirect framings like “don't aim for a perfect score” or “you may guess” diluted the effect. These findings highlight constraint-based prompting as the most reliable strategy, and we argue that constraint directives provide an actionable design pathway for practitioners seeking to integrate LLMs into effective LbT contexts.

  PublisherDOI

• Teacher Characteristics Shape Engagement and Outcomes in Online Professional Learning Environments

  2026-04-25

  articleOpen access

  This study investigates how certain teacher characteristics—specifically, math anxiety and confidence in teaching mathematics—and school-context features are associated with teachers’ behavioral engagement patterns in an online teacher professional learning platform. To this end, we applied frequent sequential pattern mining to elementary teachers’ log data collected from an online professional learning platform, the Virtual Learning Community (VLC), and linked it with survey data. Results indicate that teachers with higher levels of math anxiety were significantly more likely to remain within a single section of VLC rather than navigate across multiple sections (b = -0.764, p <.001). Additionally, this exploratory engagement was positively associated with teachers’ self-reported instructional practices (b = 0.743, p <.001). This finding indicates that teachers who navigated across multiple sections of VLC were more likely to perceive improvements in their instructional practice. Our research contributes to empirical evidence on how individual differences contribute to diverse patterns of participation in online professional learning, and it discusses practical implications that offer insights for designing teacher-specific support strategies in these environments.

  PublisherDOI

• Mobile Health for Alcohol Use Assessment: Longitudinal Effects of Breathalyzer Self-Monitoring in Everyday Contexts

  American Journal of Psychiatry · 2026-02-04 · 2 citations

  articleOpen accessSenior author

  OBJECTIVE: Although mobile health-tracking technologies have burgeoned, offering objective health information to consumers on an unprecedented scale, opportunities to directly test effects of such monitoring have been limited. Low-cost mobile breathalyzers are one tool commonly employed for blood alcohol concentration (BAC) assessment. The authors explored outcomes linked with BAC-tracking technologies, examining effects on alcohol use and self-estimation of BAC levels in a large U.S. sample. METHODS: Participants (N=32,179) were individuals who voluntarily purchased a mobile breathalyzer and provided at least three ad-lib readings between 2016 and 2022. A paired smartphone application prompted users to enter a BAC self-estimate (a guess) before the measured BAC level was displayed. Analyses included observations collected during active consumption (BAC >0.00%) from breathalyzer users who opted to share anonymized data. Breathalyzer users who displayed inattentive patterns of guessing were excluded from self-estimation analyses. The final dataset comprised 787,393 BAC readings and 387,643 self-estimates. RESULTS: The accuracy of BAC guesses increased by 2.38% over the course of breathalyzer use. Associations between breathalyzer use and BAC levels varied significantly according to participants' initial drinking levels (b=-0.0062, 95% CI=-0.0065, -0.0059). Among heavy-drinking participants, BAC levels decreased on average from 0.106% to 0.096%, whereas the reverse trend was observed for lighter-drinking participants, whose levels increased from 0.058% to 0.067%. A similar interaction emerged for BAC underestimation (b=-0.0058, 95% CI=-0.0066, -0.0049), with odds of underestimation decreasing among heavy-drinking and increasing among light-drinking participants. CONCLUSIONS: The results indicate promise for mobile BAC-tracking technologies as a low-impact intervention with the potential to decrease drinking among individuals who drink heavily-a population particularly susceptible to alcohol-related problems. In contrast, inverted trends emerged for light-drinking individuals, highlighting the need for empirical research in the fast-moving landscape of digital health.

  PublisherOA PDFDOI

• Calibration Discrepancy Predicts Students’ Subsequent Metacognitive Strategy Use in Computer-based Learning Environments

  International Journal of Artificial Intelligence in Education · 2025-10-19

  articleOpen accessSenior author

  Abstract Students often misjudge their understanding of learning material, which can lead to the use of ineffective learning strategies and result in suboptimal learning outcomes. However, it remains unclear how misjudgments relate to the use of metacognitive strategies in online learning settings, which is essential context for developing effective interventions that support students in making (and using) accurate judgments of their performance. To address this, we analyze data from 210 college students using a computer-based learning environment, investigating the relationships among calibration discrepancy, judgments, and strategies, as well as the factors affecting shifts in metacognitive judgments during learning. Students who overestimated their pretest retrospective judgments engaged less in metacognitive strategies, particularly in preparatory actions before quizzes ( b = -9.100, p &lt; .001). Notably, pretest retrospective judgments—rather than actual pretest scores—significantly predicted students’ engagement in these metacognitive strategies ( b = -9.841, p &lt; .001). Furthermore, increased engagement in repeated quiz-taking was a significant negative predictor of changes in metacognitive judgments ( b = -1.792, p = .036), indicating that students engaging in repeated quizzes tended to adjust their judgments more conservatively. These results highlight the role of pretest retrospective judgments in shaping engagement with metacognitive strategies, underscoring the importance of correcting early calibration discrepancies. Our findings advocate for early, proactive metacognitive support tools that go beyond merely presenting information, offering guidance on interpreting feedback, and implementing strategies to better align students’ judgments with their actual performance.

  PublisherOA PDFDOI

• Objective Assessment in Clinical Psychological Science: Progress in Wearable Alcohol Biosensors

  2025-10-24

  articleOpen accessSenior author

  Clinical psychology is a discipline reliant on self-reports, but uniquely susceptible to specific biases associated therewith. Here we review progress in objective behavioral assessment in the domain of alcohol research, introducing an emerging transdermal class of wearable alcohol biosensor. We note challenges of transdermal assessment, together with recent performance gains from updated devices and analytic tools, including machine learning. We indicate unanswered questions for transdermal technology, including whether devices might ultimately produce fine-grained drinking quantity estimates and device longevity. We further identify factors that can impede development of new objective measures, including the tendency to judge new tools against an implicit ideal and consider scientific findings divorced from methodological details. Finally, in evaluating transdermal and other objective measurement tools, we argue for consideration of the specific error type (random vs systematic) generally linked with novel vs existing tools, identifying measurement diversification as a priority for clinical psychology moving forward.

  PublisherDOI

• Understanding involuntary thought and affect through big data and AI.

  Technology Mind and Behavior · 2025-01-01

  articleOpen accessSenior author

  Involuntary and undirected cognitive processes, such as mind-wandering, task-unrelated thought, and spontaneous affective states, have gained significant attention in recent years. These internal states, which arise without deliberate intent, play essential roles in areas like education, mental health, and creativity but can be challenging to study due to their inherently covert nature. This special issue explores how artificial intelligence and large data sets can be leveraged to better understand these processes, offering new methods for measurement, theory development, and practical application. By integrating computational tools such as natural language processing and machine learning with cognitive theories, the contributions in this issue demonstrate how interdisciplinary approaches can uncover the dynamics of involuntary cognition and its impacts. We hope this collection inspires further research to expand the study and application of involuntary cognition.

  PublisherOA PDFDOI

• Eye-movement indices of reading while debugging Python source code

  Journal of Cognitive Psychology · 2025-01-02 · 18 citations

  article
  PublisherDOI

Recent grants

• Collaborative Research: Exploring Algorithmic Fairness and Potential Bias in K-12 Mathematics Adaptive Learning

  NSF · $987k · 2020–2025

Frequent coauthors

• Sidney K. D’Mello

  University of Colorado System

  32 shared

• Luc Paquette

  24 shared

• Jaclyn Ocumpaugh

  University of Pennsylvania

  21 shared

• Ryan S. Baker

  21 shared

• Michelle Perry

  University of Illinois Urbana-Champaign

  14 shared

• Caitlin Mills

  Minnesota Department of Education

  12 shared

• Gautam Biswas

  Vanderbilt University

  11 shared

• Stephen Hutt

  University of Denver

  9 shared

Education

• Ph.D., Computer Science

  University of Notre Dame

  2017

Awards & honors

• Best Paper Honorable Mention: 2025 ACM CHI conference on Hum…