A Comparative Study of Baseline Models for Graduation Delay Prediction Using Admission and Academic Data

Timely graduation is a crucial indicator of academic success in higher education, with broad implications for students, institutions, and stakeholders. Thus, predicting graduation delays has become an essential focus in academic management. Various studies have investigated the use of demographic and academic variables such as grade point average, gender, exam scores, and social engagement as predictors of student outcomes. In addition, various machine learning models have been applied to this task. However, identifying the best model performance for predicting student graduation delay remains a challenge. This study conducts a comparative evaluation of five baseline models-naive Bayes, decision tree, random forest, k-nearest neighbors, and logistic regression-for predicting student graduation delays. This study aims to identify the best model performance in anticipating student graduation delay. The data used in this study is internal and comes from one State Islamic University. Meanwhile, the tools used are AI Studio version 2024.1. The results of this study, the findings suggest that naive Bayes is the best model, most practical and robust for early detection of student graduation delay, especially when feature availability or data quality varies. The performance achievement of the naïve Bayes model applied to the PBS dataset is 85.71% accuracy and 0.967 AUC. While on the AS dataset, the accuracy is 90.00% and 0.875 AUC. While on the ES dataset, the accuracy is 80.00% and 0.712 AUC. These insights can support academic institutions in designing proactive interventions, helping to reduce the number of students identified as experiencing delayed graduation.