A Better Set of Object-Oriented Design Metrics for Within-Project Defect Prediction

Background: Using design metrics to predict fault-prone elements of a software design can help to focus attention on classes that need redesign and more extensive testing. However, some design metrics have been pointed out to be theoretically invalid, and the usefulness of some metrics is questioned. Aim: To identify a set of object-oriented metrics that are theoretically valid, and useful for identifying fault-prone classes in a design. Method: Drawing on four well-known sets of design metrics (CK, LK, MOOD and QMOOD), we propose a consolidated set of metrics that covers many aspects of object-oriented software design. We conduct two experiments, first using a single large system and then considering successive releases of that system, to compare the usefulness of the consolidated set with the other four sets for within-project prediction of fault-prone classes. Results: Both experiments suggest the consolidated set is effective at identifying fault-prone classes, outperforming the other metric sets (though at a cost of more false alarms). Conclusion: This paper adds to knowledge about the usefulness of existing sets of design metrics for within-project defect prediction, and identifies a consolidated set of metrics that is more effective than the existing sets at identifying fault-prone classes.