Yet another least squares estimation in NHPP-based software reliability models with grouped data
Jingchi Wu, Koutaro Daido, Tadashi Dohi, Hiroyuki Okamura · IET conference proceedings. · 2025
The present paper concerns yet another least squares estimation in nonhomogeneous Poisson process (NHPP)-based software reliability models. As Ishii et al. (2012) pointed out in the analysis of software fault-count time-domain data, since the ordinary least squares (OLS) estimation is irrelevant to the probability law of the underlying NHPP, its applicability to estimate the model parameters as well as the quantitative software reliability is rather questionable. We consider similar but somewhat different least squares estimation methods for the software fault-count time-interval data, called the group data, and compare their predictive performances with the OLS estimation and the common maximum likelihood (ML) estimation. In numerical illustrations with eleven parametric NHPP-based software reliability models, it is shown that both the OLS estimation and the ML estimation could not always provide the best predictors.