S-shaped software reliability growth models with four types of software error data

Abstract

In general, we should apply the most suitably statistical inferences to observed data. In this paper, our interests are statistical inferences on the observed data of software reliability growth models described by a non-homogeneous Poisson process having an S-shaped mean value function. We should consider such observed data in terms of censoring and removing. Censoring occurs when the test is stopped at detection of a specified number of errors or a specified time. Removing occurs when a specified number of detected errors or a total number of detected errors up to a specified time are removed. Classifying the types of software error data by censoring and removing, we can consider four types of the observed data. We give the maximum likelihood estimates of parameters in the models for four types of the observed data. The asymptotic distributions of the parameters are also discussed.