Algorithmic Prediction of Students On-Time Graduation from the University

Abstract

This study develops statistical learning models to assess the probability of undergraduate students graduating within a predetermined period, utilizing admission, performance, and demographic data. The urgency of addressing student attrition is highlighted by recent data from the National Center for Education Statistics (NCES), indicating a 59% completion rate by full-time undergraduates within six years. This research leverages institutional data from a Saudi University, focusing on freshmen enrolled in the 2012-2013 and 2013-2014 academic years, to identify students at risk of dropping out, thereby enabling timely interventions. Ten algorithms, including decision trees, ensemble models, SVM, and ANN, were built and evaluated on a test set representing 33.3% of the entire dataset using precision, recall, accuracy, and Matthews correlation coefficient (MCC). The findings show that SVM and Random Forest models were the most reliable, achieving accuracies of 0.830 and 0.831 respectively, and maintaining balance in precision, recall, and MCC. Conversely, the naïve Bayes model recorded the worst performance. The comparative analysis revealed the superior performance of ensemble models over decision tree models in predicting student attrition, emphasizing the importance of model selection in developing effective early intervention strategies. In addition, our analysis revealed that academic data is a better predictor of on-time graduation than admission data, emphasizing the need for institutions to focus on continuous academic assessment data.