BiLSTM model predicts distance-education students' academic failure risk by Week 6 from LMS weekly interaction logs, reporting 91.4% accuracy and 92.6% recall

The study proposes and evaluates a BiLSTM-based deep learning model that predicts student academic performance and failure risk from weekly interaction patterns within an LMS. Rather than relying on static or non-temporal features, the work treats time sensitivity and sequence data processing in remote learning as the central problem and uses a bidirectional long short-term memory architecture to model weekly interaction sequences. The abstract reports the model architecture (BiLSTM), the data scale (1,250 distance education students over one semester), the feature types (material access frequency, forum participation, assignment submission timing), and three classification metrics.

The model produces predictions as early as Week 6 of the course, reporting 91.4% accuracy, 89.2% precision, and 92.6% recall. This result moves the usable warning point to an early stage of the semester, giving educators an opportunity to act before the course ends. The evidence comes from experimental results reported in the abstract, including three classification metrics and an explicit time point (Week 6); the abstract does not provide baseline comparisons, cross-validation, or confidence intervals.

The study links early prediction to the goal of reducing dropout rates, pointing toward timely pedagogical intervention in digital and distance learning environments. The work connects sequence modeling results in learning analytics to an educational practice setting, emphasizing that predictions can be used to trigger instructional intervention. The abstract states that early prediction 'allows educators to implement timely pedagogical interventions to reduce dropout rates,' which is a statement of application significance; the abstract does not report outcome data from interventions actually carried out.

The result is aimed at digital and distance education settings and applies to teaching platforms and course administrators that continuously collect LMS interaction logs (material access frequency, forum participation, assignment submission timing) and can organize sequence data on a weekly basis; its design intent is to give educators a risk signal early in the course (reported as Week 6) so that timely pedagogical interventions can be arranged. For distance education institutions seeking to build early-warning systems, this work offers a modeling approach that takes weekly interaction sequences as input.

The abstract does not state which institutions or courses the data came from, how failure risk was defined and labeled, or how training and test sets were split, and it reports no comparison with baseline methods such as logistic regression or random forests, so the reference level against which 91.4% accuracy, 89.2% precision, and 92.6% recall should be read remains to be confirmed. The abstract states that early prediction can support pedagogical intervention to reduce dropout rates, but reports no outcomes from interventions actually carried out, leaving how a prediction signal translates into teaching action an open question. In addition, the loaded text is incomplete in scope, containing only the abstract and reference list, so figures, ablation experiments, and hyperparameter settings in the full text cannot be verified from the available material.