Exploiting Module Locality to Improve Software Fault Prediction

Receiving bug reports, developers usually need to spend significant amount of time resolving where to fix the faults. Although previous studies have shown that the revision frequency of a file location is an important measure to reflect the possibility of containing bugs, the frequency-based approaches achieve limited prediction accuracy for file locations having low revision frequencies. Our empirical observations show that the files of low revision frequencies in the same file directory or package of the files of high revision frequencies may be potential bug-fixing candidates for future bug reports. In this paper, we present a novel enhancement by exploiting module locality to improve the frequency-based approaches. Our experiments on three open source projects reveal that module locality can be employed to consistently improve the hit rate of a frequency-based approach and achieve the highest improvement of about 14%.