Locating where faults will be [software testing]

Abstract

The goal of this research is to allow software developers and testers to become aware of which files in the next release of a large software system are likely to contain the largest numbers of faults or the highest fault densities in the next release, thereby allowing testers to focus their efforts on the most fault-prone files. This is done by developing a negative binomial regression model to help predict characteristics of new releases of a software system, based on information collected about prior releases and the new release under development. The same prediction model was also used to allow a tester to select the files of a new release that collectively contain any desired percentage of the faults. The benefit of being able to make these sorts of predictions accurately should be clear: if we know where to look for bugs, we should be able to target our testing efforts there and, as a result, find problems more quickly and therefore more economically. Two case studies using large industrial software systems are summarized. The first study used seventeen consecutive releases of a large inventory system, representing more than four years of field exposure. The second study used nine releases of a service provisioning system with two years of field experience.