Academic performance and graduation time predictions for students from vulnerable insurgency zone using machine learning algorithms

Abstract

The timeliness of graduation is a critical performance indicator for the quality and effectiveness of higher education systems. Students enrolled into universities aiming to lay the foundation in pursuance of a decent career. Although large numbers of students enrol in universities each year, many drops out or fail to graduate within stipulated period. Understanding student’s graduation pattern can better assist universities lecturers and managers to serve student populations better in achieving their learning objectives. In the North-Eastern Nigerian context, this issue extends beyond institutional efficiency and represents a socio-technical challenge. Insurgency-related disruptions introduce unique stressors, necessitating data-driven monitoring and intervention strategies. An appropriate step is establishing a mechanism capable of monitoring and predicting graduation timelines correctly. Therefore, this study evaluates the academic performance and graduation likelihood of 1266 undergraduate students from insurgency-prone areas (IPA) at Federal University Gashua. By applying Random Forest, Logistic Regression, SVM, and ANN across imbalanced, SMOTE, and ADASYN-balanced datasets, the analysis shows that Random Forest (RF) consistently outperformed all other models with 91% precision and 97% AUC for imbalance data, 92% precision and 98% AUC for SMOTE-balance data, and 92% precision and 99% AUC for ADASYN-balance data. The results further showed that CGPA is the strongest predictor of timely graduation, while IPA status is the most influential factor associated with the likelihood of delayed graduation. Applying prediction models to timely graduation have diverse implications for both students and management thereby; harnessing preventive solution to those with likelihood of not graduating on time is of the essence.