Bayu Priyambadha¹, Tetsuro Katayama¹, Yoshihiro Kita², Hisaaki Yamaba¹, Kentaro Aburada¹, Naonobu Okazaki^1
1University of Miyazaki, 1-1 Gakuen-kibanadai nishi, Miyazaki, 889-2192, Japan
²Department of Information Security, Faculty of Information Systems, Siebold Campus, University of Nagasaki, 1-1-1 Manabino, Nagayo-cho, Nishi-Sonogi-gun, Nagasaki, 851-2195, Japan
pp. 43–48
ABSTRACT
Measuring the quality of software design artifacts is difficult due to the limitation of information in the design phase. The class diagram is one of the design artifacts produced during the design phase. The syntactic and semantic information in the class is essential to consider in the measurement process. Smell detection uses class-related information to detect the smell as an indicator of a lack of quality. Several classifiers use all information related to the class to prove how informative it for the smell detection process. The smell types that are a concern in this research are Blob and Feature Envy. The experiment using three classifiers (j48, Multi-Layer Perceptron, and Naïve Bayes) confirms that Blob smell detection utilizes the information successfully. On the other hand, Feature Envy still needs more elaboration. The average true positive rate by the classifiers is about 80.67%.
Keywords: Smell Detection, Class Diagram Smell. Design Quality, Software Design

ARTICLE INFO
Article History
Received 25 November 2020
Accepted 11 November 2021


Download article (PDF)