<![CDATA[Artificial intelligence (AI) tools are increasingly embedded in higher education, yet their relationship with students’ academic stress remains insufficiently established. This study developed and internally evaluated machine learning (ML) models for classifying academic stress vulnerability among 441 undergraduate students enrolled in AI-integrated courses at three Indonesian public universities. Using a quantitative cross-sectional predictive design, data were collected through psychometric scales, institutional GPA records, and LMS behavioral indicators. Four supervised classifiers, Gradient Boosting (GB), Random Forest (RF), Support Vector Machine with radial basis kernel (SVM-RBF), and Logistic Regression (LR), were compared using stratified train-test evaluation and five-fold cross-validation within the training data. GB achieved the strongest held-out performance (accuracy = .846, macro-F1 = .840, AUC-ROC = .930). Permutation importance indicated that cognitive load, AI literacy, and digital fatigue contributed most to classification performance. Subgroup AUC comparisons showed no significant differences across gender and discipline, although this should be interpreted cautiously. External validation and ethical governance are required before operational deployment.]]>
Copyrights © 2026
