In the realm of education, timely graduation stands as a pivotal indicator in evaluating the effectiveness of the education system. However, predicting timely graduation remains a challenge due to numerous influencing and interacting factors. In this study, researchers utilized machine learning methods to construct a predictive model capable of identifying students at high risk of not graduating on time based on their academic performance data. The dataset utilized encompassed information such as subject grades and the Grade Point Average from their previous academic records. The researchers proposed employing classification methods in the dataset prediction process, including Decision Tree (Tree), Random Forest (RF), Naive Bayes (NB), and Artificial Neural Network (ANN). To optimize dataset processing, they employed the Synthetic Minority Oversampling Technique (SMOTE) to handle imbalanced datasets. This approach yielded the best results when the original imbalanced dataset was transformed using the SMOTE technique. It was found that the RF method exhibited a dominant advantage over other methods, delivering the highest accuracy score of 99.3%, surpassing the Tree method by 0.2%. In conclusion, it can be inferred that the RF classification model, coupled with the SMOTE technique, can be optimally applied in predicting timely graduation.