Identifying Threshold Values of Change-Prone Modules
Raed Shatnawi · 2018
Software changes frequently during the lifetime of a project. These changes increase the total cost of software development. Knowing where changes are more likely help in allocating appropriate resources in different activities of software lifecycle. The change-proneness of modules is defined as likelihood of change in a module. Change-proneness can be predicted using software metrics such as Chidamber and Kemerer metrics. However, the use of prediction models requires knowledge in advanced techniques such as data mining and regression analysis. Hence, there is a need to a more direct and simple technique to have more information about change-prone modules. Threshold values are proposed as such technique to identify the modules that are more change-prone. We propose the ROC analysis to identify thresholds. The ROC analysis suggests that large values of software metrics are indicators of change proneness. The research could identify practical thresholds for four metrics, CBO, RFC, WMC and the LCOM. The inheritance metrics do not have practical thresholds.In this paper, we describe the formatting guidelines for ACM SIG Proceedings.