A Context-Aware Mobile Learning Framework for Personalized Student Support Using Artificial Intelligence

Context-aware mobile learning aims to adapt educational content not only to what a learner knows but to the situation in which learning occurs the time available, network connectivity, and time of day. Most deployed mobile-learning applications still personalize on mastery alone and ignore this situational layer, limiting real-world relevance of recommendations. This paper presents the design and implementation of an Android-based, AI-driven context-aware mobile learning framework built around a learner-model service, a context-acquisition module, and a novel context-aware linear contextual-bandit recommendation engine that jointly weighs mastery fit, time-budget fit, and connectivity fit when selecting learning activities, in contrast to prior educational bandit and collaborativefiltering approaches that condition on mastery state alone. We detail the system architecture, database schema, and recommendation algorithm with full mathematical specification, and we specify a pre-registered ten-week quasi-experimental classroom evaluation protocol. To validate the recommendation algorithm prior to classroom deployment, we conduct a controlled simulation with synthetic learners, a synthetic content catalog, and situational context sampled across thirty independent runs, extended with comparisons against established contextual-bandit baselines, feature-removal ablations, hyperparameter sensitivity analysis, and non-parametric statistical validation. The context-aware bandit achieved a mean final-window recommendation accuracy of 22.0% against a ground-truth fit criterion, significantly higher than a context-agnostic, mastery-only ablation (20.3%) and a random baseline (12.8%; paired t-tests, p < .001 for both comparisons). These simulation-based results provide initial algorithmic evidence supporting the feasibility of the proposed recommendation framework prior to real-world classroom deployment, which is specified as future work in the evaluation protocol. We conclude with a discussion of design implications, a comparison against related LLM- and bandit-based educational systems, limitations, ethical considerations, and a concrete plan for human-subject evaluation.