Predictive Modeling of Undergraduate Academic Performance Using XGBoost and Implications for Educational Policy in Nigeria

Abstract

Aim of the Study: Previous studies have paid little attention on the study that investigates the application of the XGBoost machine learning algorithm in predicting undergraduate students' academic performance, specifically focusing on final degree classification outcomes in Nigerian universities.

Methodology: Drawing on a dataset comprising academic, socio-economic, and institutional variables from 1,200 students, the study developed and validated a predictive model using supervised learning techniques. The model achieved an accuracy of 43.0%, an F1 score of 0.407, and a ROC AUC of 0.642, indicating fair discriminatory power but moderate overall predictive performance. This study therefore employed a quantitative, predictive research design using machine learning techniques to model and forecast undergraduate academic performance outcomes in Nigerian universities.

Findings: According to the reuslts, it was discovered that while XGBoost shows potential for identifying performance trends and at-risk students, its effectiveness is constrained by limitations in data diversity, class imbalance, and contextual complexity.

Conclusion: The study contributes to the emerging field of educational data mining in sub-Saharan Africa and highlights the importance of integrating machine learning tools into academic planning and decision-making. It recommends improvements in data collection practices, the inclusion of behavioural and contextual variables, and the adoption of hybrid modeling approaches to enhance accuracy.