Predicting reliability by severity and priority of defects

Quality of software systems is continuing to be an important investigation of software systems. Assessing and predicting quality attributes of object-oriented design are performed by using software metrics, knowing that a good internal structure of software system influences in a great extent its external quality attributes. This study presents an empirical investigation of software reliability. The goal is to identify the applicability of object-oriented design metrics for reliability prediction. Firstly, an estimation of the reliability is conducted. We proposed a new reliability metric at the class level considering two perspectives related to failures/bugs found, i.e. priority and severity. Later, the estimated reliability value helps us to predict the reliability of other software projects based on their internal structure. The prediction value for reliability can be made earlier in the software development life cycle. The approach’s methodology for prediction is a statistical method, the multiple linear regression considering as dependent variable for our analysis the bugs count for a class (reflected in the newly proposed metric) and as independent variables the values of the Chidamber and Kemerer (CK) metrics. The results indicated that the most influential CK metrics in predicting reliability are WMC (Weighted Methods per Class) and CBO (Coupling Between Object classes), and that the RFC (Response For Class) and LCOM (Lack of Cohesion of Methods) metrics have no impact on the value of reliability. The root mean square error is used to validate our proposed regression equation considering data from the other four projects.