What can fault prediction do for YOU?

Abstract

It would obviously be very valuable to know in advance which files in the next release of a large software system are most likely to contain the largest numbers of faults. This is true whether the goal is to validate the system by testing or formally verifying it, or by using some hybrid approach. To accomplish this, we developed negative binomial regression models and used them to predict the expected number of faults in each file of the next release of a system. The predictions are based on code characteristics and fault and modification history data. This paper discusses what we have learned from applying the model to several large industrial systems, each with multiple years of field exposure. It also discusses our success in making accurate predictions and some of the issues that had to be considered. © 2008 Springer-Verlag Berlin Heidelberg.