Performance Comparison of Bayesian Estimations on the Residual Number of Software Bugs
Yuki Hagihara, Tadashi Dohi, Hiroyuki Okamura · 2024
In this paper, we consider the posterior distributions of the residual number of software bugs in a heterogeneous testing environment with different bug detection probabilities on each testing day, and compare two prior distributions; Poisson prior and negative binomial prior, where the former corresponds to the common non-homogeneous Poisson process (NHPP)-based software reliability model (SRM), the latter to the non-homogeneous mixed Poisson process (NHMPP)-based SRM. Throughout numerical experiments with an actual software bug count data, it is shown that the common NHPP-based SRM with the Poisson prior outperformed the NHMPP-based SRM with the negative binomial prior in terms of the accurate prediction of the residual number of software bugs.