Automatic Classification of UML Sequence Diagrams from Images

Abstract

Academia’s lack of UML artifacts has been an impediment in researching UML and its implication in software development. This has initiated the conception of the UML repository, which is a platform were researchers can share and study UML artifacts. To build such a repository it’s required to collect UML diagrams. Therefore, an artifact that can automatically classify such diagrams would be of great value. In 2014, two students of University of Gothenburg successfully developed such an artifact. However, it was limited for classification of class diagrams only. This paper presents an extension of that work by including sequence diagrams, and considering that the most accurate machine learning model in the study was support vector machines, it was decided that further emphasis has to be put on researching support vector machines to maximize its usage to further improve the classifying accuracy. The data elements (feature variables) inputted to the classifier were acquired from the extracted features using image processing. The research was carried out by using a design science approach, which is an iterative research methodology that dictates an evaluation at the end of the iteration.