Enhancing Students' Academic Performance Classification in E-Learning Using Hybrid Model (Random Forest and Deep Neural Network )

Abstract

This study investigates how a hybrid model that combines Random Forest (RF) and Deep Neural Networks (DNN) might improve the classification of academic performance in e-learning environments. The study makes use of sophisticated data processing methods like feature selection and normalisation, drawing on the xAPI-Edu dataset, which comprises demographic and behavioural information from 480 students. With accuracies ranging from 68% to 92%, prior research has demonstrated the efficacy of several algorithms, including XGBoost and Logistic Regression, in forecasting student performance. These studies, however, frequently encountered difficulties with multi-class categorisation, which our model resolves by separating low, medium, and high performance with a noteworthy 80% accuracy. Crucially, the study shows that tri-class data has a detrimental effect on algorithm performance, as seen by the outcomes. With an accuracy of up to 96% in binary classifications, the hybrid model demonstrates its potential to enhance educational data mining and facilitate well-informed decision-making in academic settings.