Local polynomial software reliability models and their application

Abstract

In this paper, we propose local polynomial software reliability models (SRMs), which can be categorized into a semi-parametric modeling framework. Our models belong to the common non-homogeneous Poisson process (NHPP)-based SRMs, but possess a flexible structure to approximate an arbitrary mean value function by controlling the polynomial degree. More specifically, we develop two types of local polynomial NHPP-based SRMs; finite-failure (type-I) and infinite-failure (type-II) SRMs, which are substantial extensions of the existing NHPP-based SRMs in the similar categories. We also develop two maximum likelihood estimation algorithms in both estimation and prediction phases, where the former is used for the testing period experienced in the past, and the latter for the prediction in the future. In numerical experiments with actual 8 software fault count time-interval data sets, we compare our local polynomial NHPP-based SRMs with the well-known existing parametric NHPP-based SRMs in terms of goodness-of-fit and predictive performances. Finally, it can be concluded that our local polynomial NHPP-based SRMs with lower polynomial degrees could outperform the existing NHPP-based SRMs in several cases and should be listed as candidates for the representative NHPP-based SRMs in software reliability analysis.