USING PROCESS HISTORY TO PREDICT SOFTWARE QUALITY

Mission-critical systems must use reliable software. However, assuring software reliability often entails costly and time-consuming developm ent processes. Software quality models can mitigate these costs by pre dicting module reliability early on, which lets developers focus impro vement efforts on modules that require the most attention. Many softwa re quality models use only product metrics such as lines of code or Mc Cabe cyclomatic complexity. This product focus assumes that all module s have a similar process history. For systems that evolve, this assump tion is not valid. Modules with similar product measurements may have different quality because of different development histories. For exam ple, a reused module with many changes is likely to have more faults t han a similar module with few changes. The authors have developed a qu ality model based solely on process-history variables. Their study pos its that a module's history prior to integration can help predict the likelihood of fault discovery during integration and test. Such module reliability predictions can be used to focus review integration, and testing resources on high-risk areas of a system. They report their fi ndings in a case study involving the Joint Surveillance Target Attack Radar System, an embedded, real-time military system developed by Nort hrop Grumman for the US Air Force in support of the US Army.