A Composite Stochastic Process Model for Software Reliability

Traditional models of software reliability always assume that the failure process must follow some certain classical probability distribution and they ignore the other random factors in testing process. However, this assumption is unreasonable. And it is just the fundamental cause for no high enough precision of reliability of software and no good enough adaptability. The dynamic failure behaviors of software are decomposed into two stochastic functions, which stack and compose to a composite stochastic process model. We use Geol Okumoto model (GO model) as the non-homogeneous Poisson Process and double exponential smoothing as time series analysis composing this composite stochastic process model. The experiment shows that this model improves the precision of traditional software reliability models.