An Empirical Assessment on Program Module-Order Models

Abstract Software reliability can be improved by detecting latent software defects prior to deployment. A practical approach is to target available resources toward program modules likely to have faults, i.e., low-reliability modules. An empirical investigation on program module-order models (MOM) is presented. A module-order model assists management in prioritizing verification and validation tasks toward defective modules, providing a goal-oriented software reliability enhancement. The output of a MOM is a quality-based ranking of program modules where quality is estimated by an underlying software defect prediction model. We present an empirical investigation on module-order models designed using different underlying defect prediction models. Based on case studies of various industrial systems, it is observed that there is no single fault prediction model that always results in the preferred MOM. In fact, performance of a MOM was dependent on the cutoff range of interest, i.e., how many modules can be inspected with the given resources. Consequently, a practitioner can select a MOM based on a given cutoff range of interest.