Towards Automated Grading of UML Class Diagrams with Machine Learning
Our research describes an exploratory study on the application of machine learning for the grading of UML class diagrams. Bachelor student pairs performed a software design task for learning software design with the use of UML class diagrams. After experts had manually graded the diagrams, we trained a regression model and multiple classification models. Based on the results we conclude that prediction of a 10 point grading scale can’t be done reliably. Classifying with trained data using expert consensus and a rubric comes closer to accuracy, but is still not good enough (a precision of 69%). Future work should include larger training sets and an exploration for other features.
Dave Stikkolorum conducts research in the field of software design education at Leiden Institute of Advanced Computer Science. He is especially interested in the challenges students have to face while learning software design. Software design is the construction of the blueprint of a future software application. Dave’s research involves different didactic approaches for students such as tooling, collaborative learning and peer-learning. For publications see: https://bit.ly/2Zs7Eq7