Fault detection and prediction in an open-source software project

Abstract

Software maintenance continues to be a time and resource intensive activity. Any efforts that help to address the maintenance bottleneck within the software lifecycle are welcome. One area where such e orts are useful is in the identification of the parts of the source-code of a software system that are most likely to contain faults and thus require changes. We have carried out an empirical study where we have merged information from the CVS repository and the Bugzilla database for an open-source software project to investigate whether or not parts of the source-code are faulty, the number and severity of faults and the number and types of changes associated with parts of the system. We present an analysis of this information, showing that Pareto's Law holds and we evaluate the usefulness of the Chidamber and Kemerer metrics for identifying the fault-prone classes in the system analysed.