Tail-splitting to Predict Failing Software Modules
Ram Chillarege and Peter Santhanam
17th International Symposium on Software Reliability Engineering (ISSRE 06)
Raleigh, North Carolina, USA, November 6-11, 2006
Abstract
“Tail Splitting” is a technique to identify defect prone modules with greater focus than the commonly used Pareto principle. The power lies in factoring the development process influence on the Pareto: each module is assigned a signature depending on how its membership changes in the tail as a function of the process phases. Case studies show that Tail Splitting performs far better than the Pareto in identifying a small set of modules that will fail with high probability under customer use. The algorighm performs well for different choices of the tail boundary, yielding a tuning parameter to focus the identified module set.