Leveraging machine learning and clickstream data to improve student performance prediction in virtual learning environments

Abstract

Purpose The purpose of this research is to evaluate the utility of clickstream data and machine learning algorithms in predicting student performance and enhancing online learning experiences. By leveraging clickstream data and machine learning algorithms, the study aims to predict student performance accurately, enabling timely and personalized interventions. This approach seeks to reduce high failure and dropout rates in online courses, ultimately enhancing educational outcomes and preserving the reputation of educational institutions. Design/methodology/approach This study utilizes clickstream data from the Open University Learning Analytics Data set (OULAD) to predict student performance in virtual learning environments. The approach involves extracting and organizing data into weekly and monthly interactions. Various machine learning models, including traditional methods (Logistic Regression, Naive Bayes, K-Nearest Neighbors, Random Forest, XGBoost) and advanced time-series models (LSTM-XGBoost, GRU), are employed to analyze the data. The GRU model demonstrated the highest accuracy, offering insights into student engagement and learning patterns. Findings The study reveals that integrating clickstream data with machine learning models provides a robust framework for predicting student performance in virtual learning environments. Among the methods tested, the GRU algorithm outperformed six baseline models, achieving an accuracy of 90.13%. These findings underscore the effectiveness of using advanced time-series models to monitor and improve student engagement and success rates in online education. Originality/value This research introduces a novel approach to student performance prediction by combining traditional and advanced time-series machine learning models with clickstream data. The study’s originality lies in its comprehensive analysis of both weekly and monthly student interactions, providing educators with a powerful tool for early intervention. The findings contribute to the growing body of literature on learning analytics, offering practical solutions to enhance online education’s effectiveness and reduce dropout rates.