Using Client-Based Class Cohesion Metrics to Predict Class Maintainability

Musaad Alzahrani, Saad Alqithami, Austin Melton

Abstract: It has been established that there is a strong positive correlation between class cohesion (what we call internal connection-based class cohesion) and class maintainability. Unfortunately, the potential for using internal connection-based class cohesion to predict class maintainability is limited because we cannot measure this class cohesion until the class has been designed and coded. In this paper, we define client-based class cohesion which can be measured in the design phase. Thus, we are able in the design phase to use class cohesion to predict class maintainability. Further, the potential exists that measurements of this client-based class cohesion may in some ways be more useful than measurements of internal connection-based cohesion. This may be true because in addition to involving factors internal to the class in question, external factors, i.e., client classes, are also involved. We conduct an empirical study to investigate the extent to which a client-based class cohesion metric can be used individually and in combination with other internal connection-based class cohesion metrics to predict class maintainability. The results of the study indicate that the client-based class cohesion metric is very promising and may lead to improved results.