Assessing change proneness at the architecture level: An empirical validation

Abstract

Change proneness is a characteristic of software artifacts that represents their probability to change in future. Change proneness can be assessed at different levels of granularity, ranging from classes to modules. Although change proneness can be successfully assessed at the source code level (i.e., methods and classes), it remains rather unexplored for architectures. Additionally, the methods that have been introduced at the source code level are not directly transferable to the architecture level. In this paper, we propose and empirically validate a method for assessing the change proneness of architectural modules. Assessing change proneness at the level of architectural modules requires information from two sources: (a) the history of changes in the module, as a proxy of how frequently the module itself undergoes changes; and (b) the dependencies with other modules that affect the probability of a change being propagated from one module to the other. To validate the proposed approach, we performed a case study on five open-source projects. Specifically, we compared the accuracy of the proposed approach to the use of software package metrics as assessors of modules change proneness, based on the 1061-1998 IEEE Standard. The results suggest that compared to examined metrics, the proposed method is a better assessor of change proneness. Therefore, we believe that the method and accompanying tool can effectively aid architects during software maintenance and evolution.