Enhancing student performance prediction: the role of class imbalance handling in machine learning models

Abstract The Learning Management System (LMS) is an essential tool for educational institutions that facilitates content delivery, assessments, lecture delivery, and collaboration to enhance the learning experience. This study explores the role of LMS in creating an effective learning environment to improve students’ academic performance. To achieve the main objective of this study, we utilized a dataset xAPI-Edu-Data comprising multiple factors, such as academic, psychological, and cognitive engagement. Various machine learning techniques are employed to assess the impact of engagement activities on students’ performance. Initially, a class imbalance issue identified in the dataset and addressed using SMOTE technique. In addition, other resampling strategies applied to compare the effectiveness of proposed work. The model performance evaluated and compared using different evaluation metrics before and after data enrichment. In addition, hyperparameter optimization is conducted using a grid search approach to enhance models’ accuracy. The performance of individual models such as support vector machine (0.81), logistic regression (0.80), and decision tree (0.75) enhanced using the enriched dataset. The integration of multiple base learners into an ensemble model, with random forest as the stacking learner, achieved a weighted precision of 0.83, improving from 0.60 with the original dataset. The implementation of the stacking approach with enriched dataset has identified a better result and improved accuracy by 23%. The key contribution of this study includes identifying the effectiveness of data enrichment in improving prediction accuracy. Moreover, the research highlights the role of student engagement and behavior in measuring academic performance. The proposed model can identify the factors behind low performance, allowing further actions to be taken. Based on the prediction, the educators can work on the associated factors that could be low engagement, participation, or attendance. The findings further indicate that better use of LMS by creating more engagement activities can enhance students’ learning.