The Relationship between the Use of Artificial Intelligence and the Quality of Learning in Higher Education Institutions: A Study Based on Predictive Models

Abstract

The accelerated adoption of artificial intelligence (AI) in higher education has intensified the debate on whether AI tools genuinely enhance the quality of student learning or merely displace effort. This study examines the relationship between the intensity of AI use and the quality of learning in higher education institutions (HEIs) and, using a predictive-modeling approach, identifies which factors most strongly forecast learning quality. A quantitative, non-experimental, and predictive design was applied to a simulated cohort of 1,200 undergraduate students whose data-generating process was calibrated to effect sizes and relationships reported in the recent peer-reviewed literature. Nine predictors—AI use intensity, AI literacy, self-regulated learning (SRL), pedagogical guidance, engagement, prior GPA, study hours, socioeconomic index, and age—were modeled against a composite quality-of-learning outcome. Five algorithms were benchmarked: multiple linear regression, random forest, gradient boosting, extreme gradient boosting (XGBoost), and an artificial neural network (ANN). Multiple linear regression achieved the strongest generalization (R² = 0.567; RMSE = 6.04; 10-fold cross-validated R² = 0.579), followed closely by ensemble methods. Permutation importance identified SRL (32.2%), engagement (21.3%), and prior GPA (20.1%) as the dominant predictors, whereas raw AI use intensity contributed modestly (2.5%). Crucially, a significant AI-use × pedagogical-guidance interaction (β = 0.45, p < 0.001) indicated that AI use predicts higher learning quality only when embedded in structured pedagogical scaffolding. The findings reframe AI as an amplifier of sound learning conditions rather than an autonomous driver of quality, and they provide HEIs with an interpretable early-warning framework for data-informed decision-making.