Predictive Analysis of Learner’s Performance in Online Environments with LSTM and Attention Mechanism

Abstract

Early identification and supporting at-risk learners is a key problem in digital learning environment. This paper investigates the use of deep learning methods, namely Long Short-Term Memory (LSTM) neurons with cognitive mechanisms to determine those learners that are most likely to be at risk based upon the involvement of the learners in periodic assessment as well as engagement with the learning components in online learning environments. It also accounts for the relevance of dependencies of temporal elements, which adds a degree of precision in forecasting. The findings show how advanced analysis of data can potentially improve student support strategies with online learning systems, thus ensuring the success and retention of learners in consequence. From the test, result yield information concerning the robustness of the LSTM model in predicting the learner's achievement and provides insight into factors that most importantly have an impact on prediction. That suggests the approach of LSTM with attention mechanism is effective to capture periodic behavior of the learner on virtual platforms and early predictions will be useful for administrators to design timely intervention and improve retention rates of learners.