Development of a Machine Learning Model for Predicting Student Academic Performance at UNN

Early identification of academically at-risk students enables timely intervention and support, yet Nigerian universities largely lack data-driven systems to anticipate student performance trajectories. This study develops a machine learning model for predicting student academic performance at the University of Nigeria, Nsukka (UNN), using historical enrollment, attendance, course registration, and assessment data. Three classification algorithms, including Random Forest, Gradient Boosting, and a feedforward Artificial Neural Network, were trained and compared on an anonymized dataset of 2,400 student records spanning five academic sessions. Feature engineering incorporated CGPA trend, attendance percentage, number of courses registered per session, and departmental peer performance indices. The model development pipeline was implemented in Python using Scikit-learn and Keras, with cross-validation used for performance assessment. Results indicate that the Gradient Boosting classifier achieved the highest predictive accuracy at 87.3 percent, with precision and recall for the at-risk class of 84 percent and 89 percent respectively. The model successfully identified 91 percent of students who eventually repeated a session in the held-out test set. An early warning dashboard prototype was developed to present predictions to academic advisers. The study concludes that machine learning-based early warning systems offer a viable tool for reducing academic failure rates at UNN. Recommendations include ethical data governance frameworks and integration with the Dean of Students advisory workflow.