Investigation of State Transition Model for Predicting Software Reliability

Abstract

We compare a state transition model that is developed in this paper with previously developed software reliability models with respect to prediction accuracy. Two NASA Space Shuttlesoftwarereleasesareusedtomakethecomparison.Unexpectedly,thestatetransition modeldidnotdowell,comparedtoitsopponents,inpredictingreliabilityandremainingfailures. Furthermore, a single parameter, homogeneous Poisson model we had developed, did not faire much better. Only the two parameter, non-homogeneous Poisson model produced acceptablepredictionaccuracy.Thisresultisattributedtotheabilityofthenon-homogeneous Poisson model to capture more of the variation in failure data and, thus, estimate model parameters more accurately.