Enhancing the prediction of student performance based on the machine learning XGBoost algorithm

Abstract

Performance Factors Analysis (PFA) is considered one of the most important Knowledge Tracing (KT) approaches used for constructing adaptive educational hypermedia systems. It has shown a high prediction accuracy against many other KT approaches. While, the desire to estimate more accurately the student level leads researchers to enhance PFA by inventing several advanced extensions. However, most of the proposed extensions have exclusively been improved in a pedagogical sense, as the improvements have mostly been limited to the analysis of students’ behaviour during their learning process. In contrast, Machine Learning provides many powerful methods that could be efficient to enhance, in the technical sense, the prediction of student performance. Our goal is to focus on the exploitation of Ensemble Learning methods as an extremely effective Machine Learning paradigm used to create many advanced solutions in several fields. In this sense, we propose a new PFA approach based on different models (Random Forest, AdaBoost, and XGBoost) in order to increase the predictive accuracy of student performance. Our models have been evaluated on three different datasets. The experimental results show that the scalable XGBoost has outperformed the other evaluated models and substantially improved the performance prediction compared to the original PFA algorithm.