A study on fault-proneness detection of object-oriented systems

Fault-proneness detection in object-oriented systems is an interesting area for software companies and researchers. Several hundred metrics have been defined with the aim of measuring the different aspects of object-oriented systems. Only a few of them have been validated for fault detection, and several interesting works with this view have been considered. This paper reports a research study starting from the analysis of more than 200 different object-oriented metrics extracted from the literature with the aim of identifying suitable models for the detection of the fault-proneness of classes. Such a large number of metrics allows the extraction of a subset of them in order to obtain models that can be adopted for fault-proneness detection. To this end, the whole set of metrics has been classified on the basis of the measured aspect in order to reduce them to a manageable number; then, statistical techniques were employed to produce a hybrid model comprised of 12 metrics. The work has focused on identifying models that can detect as many faulty classes as possible and, at the same time, that are based on a manageably small set of metrics. A compromise between these aspects and the classification correctness of faulty and non-faulty classes was the main challenge of the research. As a result, two models for fault-proneness class detection have been obtained and validated.