Predictive Analytics for Student Academic Performance Using Machine Learning in Nigerian Universities

Early identification of academically at-risk students is critical for improving graduation rates in Nigerian universities, where attrition rates exceed 25 percent at many institutions. This study develops and evaluates predictive analytics models for student academic performance using machine learning, applied to student data from two Nigerian universities in North Central and South West zones. A dataset of 4,800 student records covering demographic, socioeconomic, engagement, and academic performance variables is used. Six predictive models are trained and evaluated: decision tree, random forest, k-nearest neighbours, naive Bayes, gradient boosting, and an artificial neural network. AUC-ROC and precision-recall curves are used as primary evaluation metrics to account for class imbalance between at-risk and non-at-risk student groups. Gradient boosting achieves the highest AUC of 0.91 and the strongest recall for at-risk classification. SHAP explainability analysis reveals that class attendance rate, first-semester GPA, and course load are the three strongest predictors across both universities. The study evaluates the ethical implications of predictive labelling and proposes a fairness audit procedure to detect gender and socioeconomic bias in model outputs. Findings support the development of an early intervention recommendation system for Nigerian university academic support offices. Keywords: predictive analytics, student performance, machine learning, Nigeria, academic intervention