Predictive assessment of learners through initial interactions with encoding techniques in deep learning

Abstract

Recent academic research has prominently focused on predicting student achievements in online learning. However, ongoing challenges persist for teachers and researchers, primarily revolving around the selection of relevant features. To address this issue, the study aims to predict the correlation between the use of a Virtual Learning Environment (VLE) and students’ performance based solely on their initial interactions with the platform during the initial segment of the course module within the annual study sequence of the academic year 2014–2015. The assessment employed a specialized data model that integrates the chronological order of a learner’s interactions with the educational materials provided on the online learning platform throughout the sequence employing data from the Open University Learning Analytics Dataset (OULAD), which included 32,593 undergraduates in all courses. Innovative methods, such as the VLE, and Long Short-Term Memory (LSTM) neural network rooted in deep learning, were applied to ensure the prediction of student achievement. Preliminary findings demonstrated a 60% accuracy of the model using only learner interaction data within the first third of the course duration and employing various variable encoding techniques. The successful integration of the esteemed OULAD, combined with the implementation of an LSTM neural network architecture in our data model, proved to be a highly effective strategy. This approach yielded valuable perceptions facilitating the identification of crucial data points essential for predicting student success and contributing to informed decision-making in formulating strategies to assess and support student performance effectively.