<![CDATA[The Timely student graduation is one of the key indicators used to evaluate higher education performance. Early identification of students at risk of delayed graduation enables universities to implement appropriate academic interventions and improve student success rates. This study aims to develop a student graduation prediction model using the Random Forest algorithm and to optimize its hyperparameters through Bayesian Optimization. A synthetic dataset consisting of 750 student records was employed, incorporating demographic, socioeconomic, and academic variables. The research methodology included data preprocessing, one-hot encoding, stratified data splitting with an 80:20 ratio for training and testing sets, baseline Random Forest model development, Bayesian hyperparameter optimization using the Tree-structured Parzen Estimator (TPE) with 20 optimization trials, and model evaluation using accuracy, precision, recall, F1-score, receiver operating characteristic area under the curve (ROC-AUC), and average precision. Experimental results showed that the baseline Random Forest achieved an accuracy of 70.67%, precision of 75.00%, recall of 71.43%, F1-score of 73.17%, and ROC-AUC of 82.74%. Bayesian Optimization identified the optimal hyperparameter configuration consisting of 350 trees, a maximum tree depth of six, a minimum split size of seven samples, and a minimum leaf size of four samples. Although the optimized model produced identical accuracy and F1-score values, it improved the ROC-AUC to 83.66% and the average precision to 87.68%, indicating better probability discrimination between the two classes. Feature importance analysis revealed that the number of failed courses, first-semester GPA, cumulative GPA after two semesters, attendance rate, and repeated courses were the most influential predictors. These findings demonstrate that Bayesian Optimization enhances the probabilistic ranking capability of Random Forest and can support the development of an early warning system for identifying students at risk of delayed graduation. However, further validation using real-world academic data is required before practical implementation.]]>
Copyrights © 2026
