For large scale software systems with a huge number of source codes, it
may be often useful to regard approximately software fault-counting processes
observed in testing phase as continuous-state stochastic processes. In this chapter we
introduce stochastic non-counting process models to describe the fault-detection
phenomena in software testing. The time-nonhomogeneous Gaussian process and
time-nonhomogeneous gamma process-based software reliability models (SRMs)
are summarized, and are compared with the existing SRMs such as the geometric
Brownian motion and nonhomogeneous Poisson processes (NHPPs). It is shown in
numerical examples with actual software development project data that the timenonhomogeneous
gamma process-based SRMs could provide the better goodnessof-
fit and predictive performances than the existing SRMs in many cases.
Keywords: Brownian motion process, Gamma wear process, Goodness-of-fit
performance, Information criteria, Kolmogorov-Smirnov test, Mean squared error,
Nonhomogeneous Poisson process, Predictive performance, Reliability growth
modeling, Software reliability, Stochastic differential equation.
