Automatic classification of UML Class diagrams through image feature extraction and machine learning

Abstract

Unified Modeling Language (UML) Class diagrams (CD) are a large part of the software development industry in relation to design. To be able to research UML, academia needs to have access to a database of UML diagrams. For building such a database, automatic classification of UML diagrams would be very beneficial. This research is of a design nature, and focuses on investigating CD classification: what features set them apart from other similar diagrams; how these features can be extracted through image processing; and what kind of accuracy is achievable with said features, using the Support vector machine (SVM) algorithm, and comparing it to several different machine learners (ML). The extracted features that this paper proposes for classification -- in conjunction with the chosen ML -- returns, on average, over ninety percent accuracy in classifying UML Class diagrams.