BiLSTM-Based Deep Learning Model for Early Prediction of Student Academic Failure Risk in Digital and Distance Education

Abstract

Digital and distance learning platforms generate abundant interaction log data from students. However, leveraging temporal log data for early prediction of academic failure risks remains challenging due to time sensitivity and sequence data processing in remote learning contexts. This study aims to develop a Deep Learning model based on Bidirectional Long Short-Term Memory (BiLSTM) to predict student academic performance and failure risks based on weekly interaction patterns within a Learning Management System (LMS). A dataset comprising interaction logs from 1,250 distance education students over one semester was analyzed by extracting features such as material access frequency, forum participation, and assignment submission timing. Experimental results demonstrate that the BiLSTM model achieves an accuracy of 91.4%, Precision of 89.2%, and Recall of 92.6% as early as Week 6 of the course. This early prediction allows educators to implement timely pedagogical interventions to reduce dropout rates in digital and distance learning environments.