Fitting and prediction uncertainty for a software reliability model

Abstract

The cost of system operational testing is steadily increasing. It is desirable for the software manager to know if the software is sufficiently well developed or reliable to support such testing. Current software reliability models provide only estimates of the mean time to next failure or expected numbers of errors to occur in additional testing time. The goal of this thesis is to take into account prediction uncertainties of a software reliability model. Bootstrapping is used to provide the software manager with confidence limits of the predicted expected numbers of faults to occur for additional testing time. the result can be particularly useful for a software manager who has to answer a subjective question: is the software reliable enough to support system operational testing?. A range of predicted number of faults will be of more use to a software manager, who has to justify the answer to this question, than just a point estimate. Two software fault data sets are analyzed with this technique emphasizing how a software manager should analyze the results.