Yeni Anistyasari, Suparji, Ekohariadi, Shintami C Hidayati, Agus Wiyono, Awwalia Arofatun Nikmah
Digital literacy is a critical competency in higher education, yet its multidimensional nature makes it difficult to assess and model using traditional approaches. This study proposes an AI-driven framework for mining digital literacy competency patterns within adaptive learning environments. Data from 230 undergraduate computer science students were collected through surveys, LMS behavioral logs, and performance records. A deep autoencoder was used for representation learning, followed by clustering algorithms (KMeans, DBSCAN) to identify homogeneous student profiles. Supervised models (Decision Tree, Random Forest) were then applied to predict engagement and performance. Finally, explainable AI techniques (SHAP, LIME) provided interpretable insights. Results revealed three distinct profiles - Technically Skilled but Critical-Weak, Balanced Digital Users, and Critical Creators-that significantly correlated with engagement and academic outcomes. Random Forest achieved the best performance (ROC-AUC =0.86). SHAP analysis highlighted critical literacy and projective literacy as dominant predictors of success. These findings demonstrate how AI-based pattern mining can generate actionable insights for adaptive learning design, bridging computational methods and educational practice. © 2025 IEEE.
Universitas Negeri Surabaya, Faculty of Engineering, Surabaya, Indonesia; Institut Teknologi Sepuluh Nopember, Dept. of Informatics, Surabaya, Indonesia
