Software Reliability Modeling Based on Zero-truncated and/or Zero-inflated Compound Distributions
Jingchi Wu, Tadashi Dohi, Hiroyuki Okamura · 2023
Software reliability growth models (SRGMs) are used to assess quantitative software reliability and to monitor/control software testing progress. During almost the last five decades, SRGMs based on non-homogeneous Poisson processes (NHPPs) have gained much popularity for describing the stochastic behavior of the cumulative number of software faults detected in testing phase, because of their tractability and goodness-of-fit performances. Grottke and Trivedi (2005) proposed an interesting NHPP-based modeling framework, called all-stage zero-truncated NHPP-based SRGMs, and showed their goodness-of-fit and predictive performances with several actual software development project data. In this paper we further generalize their idea on all-stage zero-truncation by introducing zero-truncated and/or zero-inflated compound probability distributions. Throughout comprehensive numerical experiments, we compare our generalized modeling frameworks with the existing ones in terms of goodness-of-fit and predictive performances, and show that the zero-truncated NHPP-based SRGMs are still attractive more than the others including the non-truncated SRGMs.