Developing AI-Driven Adaptive Learning for Science Experiment Literacy

Thesis Abstract

Developing AI-Driven Adaptive Learning for Science Experiment Literacy addresses a critical gap in secondary and tertiary science education where students’ conceptual understanding of experimental design and data interpretation lags behind procedural proficiency. Despite widespread access to digital laboratories and simulations, learners often receive uneven, one-size-fits-all instruction that fails to adapt to individual prior knowledge, misconceptions, and procedural fluency. This study aims to design, implement, and evaluate an AI-driven adaptive learning (ADL) system that tailors scaffolded science experiment literacy through real-time analytics, natural language feedback, and multimodal simulations. The core objective is to determine whether adaptive, evidence-based interventions improve students’ abilities to formulate hypotheses, design robust experiments, interpret data, and draw valid conclusions more effectively than conventional instruction. The research employs a quasi-experimental mixed-methods design conducted across two phases in two urban high school science programs and one university introductory biology course, spanning a full academic year. A total of 1,200 learners will participate 800 in intervention arms (n = 400 secondary students in two schools and n = 400 university students) and 400 in control arms receiving standard instruction. The ADL system will integrate a Knowledge Tracing model to estimate evolving mastery in experimental concepts, a Bayesian Knowledge Network to map prerequisite relationships, and a reinforcement-learning–driven recommendation engine that personalizes tasks, prompts, and simulations. Data collection instruments include (i) standardized science literacy assessments administered at baseline, mid-term, and post-intervention; (ii) system-generated analytics on learners’ experiment design decisions, data interpretation accuracy, and evidence-quality scoring; (iii) validated attitudinal scales toward science and technology use; (iv) semi-structured interviews and focus groups with a purposive subsample of teachers and students to capture perceived efficacy and usability. Instrument validity will be established through content validity panels and pilot testing (n = 120). Reliability will be assessed using Cronbach’s alpha for scales and inter-rater reliability for qualitative coding (Cohen’s kappa ? .80). Quantitative data will be analyzed using hierarchical linear modeling to account for nested data (students within classes, classes within schools), with fixed effects for group (ADL vs. control), time, and interaction terms to examine differential gains. Regression analyses will assess relationships between mastery estimates from Knowledge Tracing and outcome measures of experimental literacy. Mediation analyses will explore whether improvements in procedural fluency mediate gains in conceptual understanding. Qualitative data from interviews and focus groups will undergo reflexive thematic analysis to identify themes related to perceived autonomy, feedback quality, and alignment with scientific practices. A convergent mixed-methods approach will triangulate findings to elucidate how adaptive feedback cycles influence learner engagement and conceptual advancement. Ethical considerations include informed consent, data privacy, and safeguarding vulnerable participants, with approvals from institutional review boards. Expected findings anticipate that learners in the ADL condition will exhibit statistically significant improvements in experimental literacy scores (expected effect size d ? 0.35–0.50) compared to controls, with larger gains among students with initially lower mastery levels. The study also expects enhanced epistemic agency evidenced by more accurate hypothesis formulation, robust error analysis, and higher-quality evidence in argumentation. Qualitative insights are anticipated to reveal higher perceived relevance of tasks, better alignment of prompts with learner misconceptions, and improved perceived teacher support through actionable feedback. The study contributes to knowledge by advancing a scalable, data-driven pedagogy that operationalizes adaptive learning theories within science experimentation contexts. It synthesizes Knowledge Tracing, Bayesian networks, and reinforcement learning to create a theoretically informed, practically implementable model for improving science experiment literacy. Findings will inform the design of scalable ADL frameworks for STEM education, contribute to theory on how adaptive feedback shapes scientific reasoning, and offer empirical benchmarks for measuring experimental literacy in mixed-ability cohorts. Practical recommendations will address curriculum integration, teacher professional development, and the ethical deployment of AI in classroom assessment. Overall, the research concludes that AI-driven adaptive learning can substantially elevate science experiment literacy by personalizing cognitive load, reinforcing experimental reasoning, and systematically exposing learners to progressively complex investigative tasks.