EVALUATION OF RELATIONSHIP BETWEEN OBJECT ORIENTED METRICS AND CHANGE PRONENESS

Abstract

Changes are always inevitable. As the software evolves changes might occur due to some defect or when some additional functionality is added to the software. Change proneness is the probability of changing some part of software. The requirement modification is needed if some changes occur. If there are more changes needed in the software, then this means that there is a problem of design quality and therefore it’s design needs to be improved. In such cases, it is very important to discover change prone classes in the software in early phases of software development so that testing resources can be planned to reduce the maintenance effort. As a result testing becomes more qualitative because more focus will be laid on those classes that are more prone to changes. By doing so, the probability of occurrence of defects can be reduced which can thereby lead to better maintenance. Our study analyzes the relationship between object oriented metrics and change proneness. Statistical and machine learning methods have been studied for predicting change prone classes. These methods have been applied on five open source java projects namely ABRA, ABBOT, APOLLO, AVISYNS and JMETER. The performance has been analyzed on the basis of receiver operating characteristics. Results have shown that the performance of machine learning techniques is comparable to statistical methods.