<![CDATA[Digital literacy has become an essential skill in higher education, particularly in online and distance learning settings. This study explores the use of Light Gradient Boosting Machine (LightGBM) to classify digital literacy levels among 10,393 students at Universitas Terbuka. To improve both efficiency and clarity of interpretation, feature selection was carried out using SelectKBest, which reduced the dataset to 33 predictors. The final model, evaluated through stratified 5-fold cross-validation, achieved an accuracy of 0.964 and a weighted F1-score of 0.964. The results show that limiting the number of features did not weaken predictive performance, while also making it easier to identify which aspects of digital literacy are most influential. Interestingly, the strongest predictors were not only technical skills but also ethical behavior, digital citizenship, and online communication. These findings highlight that digital literacy is multidimensional and that effective assessment tools must account for social and behavioral factors alongside technical competence. Taken together, applying feature selection with LightGBM offers an effective way to assess digital literacy in higher education. The method achieves strong predictive accuracy while keeping the model interpretable, giving universities clearer guidance for shaping interventions and curricula in online learning contexts.]]>
Copyrights © 2025
