Effects of incorporating special methods into cohesion measurement on class instantiation reuse‐proneness prediction

Abstract

The previously proposed class cohesion measures employ different approaches to assess the strength of the relations between the attributes and methods in a class. Access methods, constructors and destructors are special types of methods with special characteristics that can falsely alter the class cohesion measurement. In this study, the authors empirically explored the impact of considering special methods (SPs) on the cohesion measures’ abilities to predict the classes that can be intensively reused via instantiation (IRI). They considered classes in the JHotDraw and Eclipse systems. For each class, they obtained cohesion results using 17 measures in four different scenarios of considering or ignoring SPs. They collected the instantiation reusability data and applied a statistical technique to build a prediction model using each measure in each considered scenario. They investigated the significance of the changes in the prediction results. The authors’ results demonstrated that cohesion had a negative impact on class instantiation reuse-proneness and that SPs had significant impacts on cohesion values and the abilities of the cohesion measures to predict IRI classes. In practice, when applying cohesion measures to predict IRI classes, the results suggest that SPs must be included in cohesion measurement.