Detection of Graduation Potential in Prospective Students using the Random Forest Algorithm

Abstract

Detecting students’ graduation potential is commonly performed by evaluating various academic and non-academic factors. This study aims to develop a predictive model for student graduation from the beginning of their academic journey, utilizing high school academic data such as grades, attendance, study hours, as well as demographic and social factors. The goal is to enable universities to identify students who are at risk of delayed graduation. With accurate predictions, institutions are expected to design more targeted academic interventions, such as tutoring, counseling, or other forms of academic support. A total of 396 student records were used in this study and processed through a series of preprocessing steps, including the removal of irrelevant data and the encoding of categorical variables. The model was developed using the Random Forest algorithm with parameters set to max_depth = 15 and random_state = 42. Model performance was evaluated using accuracy, recall, F1-score, and the ROC curve. The results show that the model achieved an accuracy of 89%, with the Pass class having a recall of 87% and an F1-score of 91%, and the Fail class showing a recall of 92% and an F1-score of 84%. Additionally, the Area Under the Curve (AUC) value of 0.94 indicates excellent model performance in distinguishing between students likely to graduate and those at risk of not graduating.
This study confirms that the model is effective in classifying graduation outcomes based on early academic data. For further development, it is recommended to include additional variables such as psychological factors, learning motivation, and socioeconomic conditions. Moreover, tuning the model by adding other parameters—such as n_estimators, min_samples_split, and max_features—is suggested to improve the model’s accuracy and generalizability.