Didik Dwi Prasetya, Roudhotulloh Nazakhan, Wahyu Sakti Gunawan Irianto, Utomo Pujianto, Zulfiqar Naufal Maulana, Ahmad Kholish Fauzan Shobiry
Manual assessment of concept maps often requires considerable time and is prone to subjectivity and inconsistency across evaluators. To address this challenge, this study investigates the use of machine learning to automate the evaluation of concept-map proposition quality. The Decision Tree and Random Forest were involved in predicting proposition quality scores. A total of 691 propositions collected from student-generated concept maps on Relational Database topics were preprocessed through text cleaning and transformed into numerical features using TF-IDF. Both models were trained using an 80:20 data split, and hyperparameters were optimized using GridSearchCV. Model performance was evaluated using accuracy, precision, recall, F1-score, confusion matrix, and Cohen's Kappa as a reliability measure. The Random Forest model achieved the best performance with an accuracy of 81.2% and a Cohen's Kappa of 0.872, outperforming Decision Tree, which achieved an accuracy of 79.8% and a Kappa of 0.852. These results demonstrate that Random Forest offers higher reliability and consistency in automatic proposition assessment, indicating its potential for supporting scalable and objective evaluation of student concept maps in educational settings. © 2025 IEEE.
Universitas Negeri Malang, Department of Electrical Engineering and Informatics, Malang, Indonesia; Univeritas Negeri Malang, Department Teknik Elektro Dan Informatika, Malang, Indonesia
