Evaluating the Impact of Multi-Layer Data on Machine Learning Classifiers for Predicting Student Academic Performance

Abstract

It is important for educational institutions to be able to accurately assess students’ performances to enable them to intervene and improve the learning process. This paper assesses the effectiveness of including multi-layer data in the performance of machine learning classifiers for student success prediction. By integrating student registration data with course-level Intended Learning Outcomes (ILOs), we compare several classifiers such as Logistic Regression, Random Forest, XGBoost, Support Vector Machines (SVM), and others. It is observed that integrating course-level ILOs data enhances classifier effectiveness in all metrics. For Random Forest, the accuracy improved to 0. 844 with the ILOs included compared to 0.770 using only registration data, but all others such as XGBoost and Logistic Regression demonstrate a significant improvement too. Therefore, the employment of multi layered data in addition to enhancing the predictive power of the models, gives institutions a holistic understanding of students’ learning progression for immediate and focused interventions. These results reinforce our need to use rich data to improve the predictive power of academic achievement and student services.