<![CDATA[Reading interest is an important indicator of academic engagement and lifelong learning, yet its relationship with digital learning behavior remains underexplored in higher education. This study proposes an explainable machine learning framework to predict student reading interest using digital learning behavior, reading habits, and demographic characteristics. Data were collected through self-reported questionnaires from 300 Indonesian university students. Reading interest was formulated as a binary target variable, and three feature scenarios were evaluated: digital behavior only (S1), reading behavior only (S2), and an integrated model combining all features (S3). The dataset was divided using an 80:20 stratified train–test split, with SMOTE applied only to the training data to address class imbalance while preventing information leakage. S3 achieved the highest ROC-AUC (91.92%) and cross-validation F1-score (86.62% ± 3.34%), while S1 obtained the same test accuracy (84.29%) and a slightly higher test F1-score (85.33%). SHAP analysis provided interpretability by identifying daily reading duration and monthly book count as key predictors. These findings indicate that multidimensional behavioral indicators can support early student-engagement screening systems. However, results from a single institutional context require cautious interpretation.]]>
Copyrights © 2026
