Adaptive Intelligent Tutoring System for Self-Paced Computer Science Education

Thesis Abstract

This study investigates the effectiveness of an Adaptive Intelligent Tutoring System (AITS) designed to support self-paced learning in computer science by delivering personalized pedagogy, real-time assessment, and targeted feedback to learners across diverse backgrounds. The research addresses the problem of plateaued achievement and uneven progression in self-directed CS education due to variable prior knowledge, motivation, and access to expert scaffolding. The aim is to evaluate whether an AITS that integrates mastery learning, problem-based practice, and transparent progress dashboards can improve learning outcomes, engagement, and self-regulation for undergraduate and early graduate CS students. Specific objectives are (1) to design an AITS architecture incorporating student modeling, adaptive content sequencing, and automated formative assessment grounded in constructivist and cognitive apprenticeship principles; (2) to determine the system’s effect on knowledge gains in key CS topics (programming fundamentals, data structures, and algorithms) compared with a non-adaptive online module; (3) to examine changes in learner motivation, perceived self-efficacy, and metacognitive awareness; (4) to analyze system usability, user experience, and engagement metrics; and (5) to identify contextual factors that influence adaptive tutoring efficacy across demographic subgroups. The methodological approach adopts a quasi-experimental mixed-methods design over a full academic term (12 weeks) with two parallel cohorts an intervention group (n = 180) using the AITS and a control group (n = 180) using standard online resources. The population comprises first- and second-year undergraduate computer science students enrolled in introductory and intermediate CS courses at a large public university. Data collection instruments include standardized pre- and post-tests aligned with ACM CS curriculum outcomes, an instrument measuring intrinsic motivation and self-efficacy in CS (validated for university students), system usage analytics (time-on-task, hint requests, mastery progress), a usability scale (SUS), and semi-structured interviews with a purposive subsample (n = 24) to explore experiential insights. The AITS employs a Bayesian knowledge tracing model for student proficiency estimation, a mastery-based progression algorithm, and reinforcement learning–driven content sequencing to optimize problem difficulty. Data analysis integrates quantitative and qualitative procedures descriptive statistics and normality checks, ANCOVA and multilevel modeling to compare learning gains while controlling for prior knowledge and demographic covariates, regression analyses to identify predictors of achievement and engagement, and thematic analysis of interview transcripts to extract design implications and perceived barriers. The study adheres to validity and reliability standards through instrument pilot testing (n = 60), measurement invariance checks, triangulation of performance data with usage logs, and inter-rater reliability for qualitative coding (? ? 0.80). Ethical considerations include obtaining informed consent, ensuring data anonymity, and securing institutional review board approval. Expected findings anticipate that the AITS will produce statistically significant higher post-test gains in programming fundamentals, data structures, and algorithms (effect size partial ?² ? 0.08–0.15) compared with the control condition, with greater advantages observed among students with lower prior CS exposure. It is hypothesized that adaptive feedback, timely hints, and mastery-based progression will correlate positively with increases in intrinsic motivation, CS self-efficacy, and metacognitive awareness (p < .05). Usage analytics are expected to reveal that higher engagement with scaffolded micro-challenges and frequent mastery checks mediate learning outcomes, while system usability will be a moderating factor. The qualitative findings are anticipated to converge on themes of personalized pacing, transparency of progression, perceived legitimacy of hints, and concerns related to over-reliance on automated guidance. The study contributes to knowledge by providing rigorous evidence on the pedagogical value of AITS in self-paced CS education, detailing a scalable architectural blueprint that integrates Bayesian student modeling with mastery-based sequencing and reinforcement learning for content adaptation. It advances theoretical understanding of how adaptive tutoring interacts with constructivist and cognitive apprenticeship frameworks in computer science contexts, and it informs best practices for deploying intelligent tutoring in large, diverse CS cohorts. The main conclusion is that a well-designed AITS can meaningfully enhance learning gains and motivation in self-paced CS education, particularly for learners with limited prior exposure, when coupled with robust usability and transparent progress feedback. Recommendations include refining the knowledge tracing granularity, expanding topic coverage to more advanced CS domains, implementing longitudinal follow-ups to assess knowledge transfer, and exploring cross-institutional deployments to validate generalizability.