Deep learning–driven learning analytics for enhancing digital literacy skills of media students using an explainable BiLSTM-attention framework

The increasing adoption of digital technologies and online learning platforms has transformed higher education, making digital literacy an important competency for media students. However, existing learning analytics approaches primarily focus on academic performance prediction while providing limited attention to digital-literacy-related behavioral patterns and the interpretability of predictive models. Furthermore, conventional machine learning techniques often fail to capture the temporal behavioural patterns generated through learner interactions within Learning Management Systems (LMSs). To address these challenges, this study proposes an Explainable BiLSTM-Attention Learning Analytics Framework for predicting MDLA-derived digital-literacy behavioral proxy categories using learner behavioural data extracted from the Open University Learning Analytics Dataset (OULAD). Because OULAD does not contain independently measured digital-literacy scores, the Low, Medium, and High target categories are derived from LMS behavioral information through the proposed Media Digital Literacy Analytics (MDLA) representation and are treated as behavioral proxy categories rather than directly measured levels of digital literacy. The proposed framework combines a Bidirectional Long Short-Term Memory (BiLSTM) network with an attention mechanism to model sequential learner behaviour, while SHapley Additive exPlanations (SHAP) and attention visualization improve model transparency and interpretability. Experimental results indicate that the proposed framework obtained higher scores than the evaluated baselines in a single run on OULAD, the evaluated machine learning and deep learning baseline models, achieving an Accuracy of 95.8%, Precision of 95.1%, Recall of 94.7%, F1-score of 94.9%, and a ROC-AUC of 97.2%. Explainability analysis identified engagement duration, assessment performance, resource usage, forum participation, and interaction frequency as the most influential behavioural factors contributing to the prediction of the MDLA-derived proxy categories. The findings suggest that integrating sequential deep learning with explainable learning analytics can model digital-literacy-related behavioral profiles in this OULAD setting. MDLA is a behavioral representation and has not been validated against a standardized digital-literacy assessment. The findings are based on a single dataset (OULAD) and a single experimental run, and should not be generalized to other institutions, LMS platforms, or learner populations without external validation. The findings are based on a single dataset (OULAD) and a single experimental run, and should not be generalized to other institutions, LMS platforms, or learner populations without external validation.