Correlating software metrics with software defects

In software development and testing an interesting issue is checking correlations of observed software defects with various product and process metrics. Such analysis is helpful in predicting potential defects and optimization of testing processes. In the paper we present results of deeper studies in this area, they involve many metrics and various prediction schemes taking into account diverse correlation parameters. Special attention is paid to the problem of selecting the most significant metrics. In the prediction schemes we consider modified and non modified program objects. We have created a special analysis platform allowing us to perform a wide scope of data exploration. The presented analysis methods have been verified in an experimental investigation covering twelve open source projects, for some of them several subsequent versions have been examined. This is followed by a summerizing result discussion.