Naohiko Tsuda, Hironori Washizaki, Yoshiaki Fukazawa, “Evaluating Structural Validity of UML Class Diagrams by Measuring the Number of Highly Responsible Classes,” 13th IASTED International Conference on Software Engineering (SE2014), Innsbruck, Austria, February 17 – 19, 2014. (to appear) (iastedDocVer5-6eng.pdf)

Design models are often developed using UML class diagrams. In past questionnaire surveys, Lange and Nugroho reported that designers tend to write the important or complex parts of a design model in detail. Thus, we presume that a design class diagram in which some parts are more important will contain classes with both high and low responsibility. Moreover, we hypothesize that a design class diagram containing few highly responsible classes has low validity. In this study, we calculated four basic class metrics (the number of attributes (NAttr), operations, associations and subclasses) and our novel metrics (e.g. the number of larger values of NAttr). All metrics were calculated using 65 design class diagrams, which were originally submitted to a Robot Contest on the domain of embedded systems and evaluated by software development experts based on the understandability of the system, adequacy of responsibility assignment, etc. Then the relations between our metrics and the experts’ qualitative assessment were analyzed. Consequently, the usefulness of our metrics and our hypothesis are confirmed empirically.