Identifying fault-prone files using static analysis alerts through singular value decomposition
Mark S. Sherriff, Sarah Smith Heckman, Mike Lake, Laurie A. Williams · Proceedings of CASCON · 2007
Static analysis tools tend to generate more alerts than a development team can reasonably examine without some form of guidance. In this paper, we propose a technique for leveraging field failures and historical change records to determine which sets of alerts are often associated with a field failure using singular value decomposition. We performed a case study on six major components of an industrial software system at IBM over six builds spanning eighteen months of development. Our technique identified fourteen alert types that comprised sets of alerts that could identify, on average, 45% of future fault-prone files and up to 65% in some instances.