Software defect prediction based on source code metrics time series

Source code metrics have been proved to be reliable indicators of the vulnerability of the source code to bugs. Typically, a source code unit with high value of a certain metric is considered to be badly structured and thus error-prone. However, analysis of source code change history shows that there are cases when source files with low values of metrics still turn out to be buggy. Instead of introducing new metrics for such cases, I investigate the possibility of estimating the vulnerability of source code units to bugs on the basis of the history of the values of selected well-known metrics. The experiments show that we can efficiently identify bad source code units just by looking at the history of metrics, coming from only a few revisions that precede the bug fix.