Exploring the predictive performance of the generalized order statistic software reliability models

This dissertation investigates the use of Generalized Order Statistic Model formulation of the Nonhomogeneous Poisson Process in software reliability modeling. Specifically, we compare the fits for common parametric assumptions for the underlying distributional assumptions of the Generalized Order Statistic Model. Fifty one software data sets were complied from a host of sources in the literature and private communications. The data sets were used to fit the Generalized Order Statistic models using maximum likelihood estimation. Once fit, the models were compared for goodness of fit to the data set and for estimation of the total number of software faults for the data set.