Influence of Nominal Project Knowledge in the Detection of Design Smells: An Exploratory Study with God Class

Several Design Smell detection tools have been developed for identifying Design Smells in source code or design models. The early prediction of a useful set of Design Smells has a positive impact on software quality. In this paper, we present an exploratory study to check whether some project information can be relevant or not to be supplied to a classifier in order to detect God Class Design Smell. We want to explore if clarifying the domain, the status and the size category of the project to which a class belongs to can lead to variations in the classification accuracy and usefulness for this Design Smell detection. The dataset is formed by the 12786 classes of 24 projects with different size categories, domains and maturity status. We conduct the experiments with eight different machine learning approaches which are the most recently used in literature. These eight involve all families of classifiers. The results of classifiers are compared based on the accuracy, sensitivity and specificity performance significance tests. We find that the set of nominal project knowledge studied in this paper have not any impact on the detection of God Class Design Smell based on the set of detection tools were used to identify the God Class Design Smell.

keywords: Design Smell detection, God Class, Machine learning;