AI-driven fitness coaching for personalized physical education in schools

Thesis Abstract

This study addresses the escalating gap between prescribed physical education curricula and individual student needs by examining the implementation of AI-driven fitness coaching to personalize physical education in school settings. The problem centers on one-size-fits-all PE programs that fail to accommodate varying fitness levels, motivation, and learning styles, potentially limiting student engagement and long-term physical activity adherence. The aim is to evaluate whether an AI-enabled coaching system can tailor PE interventions to individual students while maintaining alignment with curricular standards. Specific objectives include (1) determining the impact of AI-driven personalized coaching on student moderate-to-vigorous physical activity (MVPA) minutes per PE lesson; (2) assessing changes in student motivation, self-efficacy, and intrinsic goal orientation; (3) analyzing the accuracy and usability of the AI coaching platform in real-world classrooms; (4) identifying teachers’ perceptions of feasibility, acceptability, and integration with existing PE pedagogy; and (5) exploring equity considerations across student subgroups (sex, socio-economic status, baseline fitness). The study employs a mixed-methods, quasi-experimental design conducted over a 12-week term in five urban secondary schools with diverse student populations (approximately 1,200 participants). A cluster randomized assignment assigns classrooms to either the AI-driven personalized coaching condition or to standard PE instruction. Quantitative data are collected through wearable accelerometers (ActiGraph GT3X+) to quantify MVPA and sedentary time, validated self-report instruments for motivation (Academic Motivation Scale adapted for PE), self-efficacy (Physical Activity Self-Efficacy Scale), and intrinsic goals (Learning and Performance Orientations Questionnaire). Instrument validity and reliability are confirmed with Cronbach’s alpha above 0.80 and test–retest reliability coefficients exceeding 0.70. Data collection occurs at baseline, mid-intervention (6 weeks), and post-intervention (12 weeks). Additionally, system logs capture AI recommendations, adherence, and engagement metrics. Qualitative data are gathered via semi-structured interviews and focus groups with 30 teachers and 60 students, analyzed using thematic analysis to extract dimensions of usability, perceived impact on pedagogy, and perceived equity. The quantitative analysis employs multilevel mixed-effects regression to account for clustering at the classroom level, with MVPA as the primary outcome and covariates including baseline fitness, gender, and school. Mediation analyses explore whether changes in motivation and self-efficacy mediate the effect of AI coaching on activity outcomes. A repeated-measures ANOVA tests temporal trends, while subgroup analyses examine differential effects across demographic strata. The qualitative data are triangulated with quantitative findings to generate a holistic interpretation. Expected findings include (a) a statistically significant increase in weekly PE MVPA minutes among students receiving AI-driven coaching compared with controls (p < 0.05), with effect sizes in the small-to-moderate range (Cohen’s d ? 0.25–0.40); (b) improvements in autonomous motivation and perceived self-efficacy mediating the relationship between AI coaching and activity levels; (c) high system usability scores (System Usability Scale ? 70) and favorable teacher attitudes toward integration with existing curricula, alongside identifying practical barriers such as device management and data privacy concerns; (d) equitable benefits across subgroups, albeit with nuanced differences linked to baseline fitness and access to technology. The study contributes to knowledge by providing empirical evidence on the feasibility, effectiveness, and equity of AI-driven personalized PE, advancing theoretical understanding of technology-enhanced motivation within physical education, and offering a validated framework for implementing adaptive coaching in schools. Theoretical underpinnings include Self-Determination Theory to explain motivational shifts and the Technology Acceptance Model to interpret adoption dynamics, augmented by the Health Belief Model to contextualize behavior change. Recommendations include (1) scale-up strategies for AI-enhanced PE with robust data governance and privacy safeguards; (2) professional development for teachers on interpreting AI-generated insights and adapting pedagogy; (3) design enhancements to ensure accessibility and inclusivity across diverse student populations; and (4) policies to align AI coaching with curricular standards while preserving student well-being and autonomy.