Predicting academic performance of students with machine learning

Abstract

This study investigates the effectiveness of machine learning and deep learning models for early prediction of student performance in higher education institutions. Using the Open University Learning Analytics (OULA) dataset, various models, including Decision Tree, Support Vector Machine, Neural Network, and Ensemble Model, were employed to predict student performance in three categories: Pass/Fail, Close to Fail, and Close to Pass. The Ensemble Model (EM) consistently outperformed other models, achieving the highest overall F1 measure, precision, recall, and accuracy. These results highlight the potential of data-driven techniques in informing educational stakeholders' decision-making processes, enabling targeted interventions, and facilitating personalized learning strategies tailored to students' needs. By identifying at-risk students early in the academic year, institutions can provide additional support to improve academic outcomes and retention rates. The study also discusses practical implications, including the development of pedagogical policies and guidelines based on early predictions, which can help educational institutions maintain strong academic outcomes and enhance their reputation for academic excellence. Future research aims to investigate the impact of individual activities on student performance and explore day-to-day student behaviors, enabling the creation of tailored pedagogical policies and guidelines.