Abstract
In this paper, we propose statistical data analysis models that enable us to assess software reliability more accurately than conventional models. These models are based on discrete analogs of NHPP (nonhomogeneous Poisson process) software reliability models which have exact solutions in terms of solving the hypothesized differential equations. The difference equations derived by transforming the continuous testing time into the discrete one, and their exact solutions are reduced to the differential equations on which the NHPP models are defined and the solutions, respectively, when the discrete time-interval tends to zero. The proposed discrete models conserve the characteristics of these NHPP models because the difference equations have exact solutions. Finally, we show numerical examples of software reliability analysis by using actual data.
