Infinite-Failure software reliability models based on non-homogeneous markov processes

Abstract

This paper develops and evaluates a new non-homogeneous Markov process (NHMP)-based framework for software reliability modeling under imperfect debugging. Existing NHMP-based software reliability models are built from bounded cumulative baseline intensities and therefore correspond to finite-failure formulations. This assumption can be restrictive when debugging may introduce new faults and the cumulative number of detectable faults is not naturally bounded. To address this issue, we derive NHMP transition rates from failure-rate baseline functions, following the modeling idea of infinite-failure NHPP-based software reliability models. Using two subclasses of NHMPs, namely generalized binomial processes (GBPs) and generalized Pólya processes (GPPs), we formulate 22 models, consisting of 11 alternative GBP-based models and 11 GPP-based models. Their goodness-of-fit and predictive performances are evaluated using 8 software fault-detection time-domain data sets and 8 grouped data sets, and are compared with representative NHPP-based software reliability models. The numerical results show that the proposed framework often improves goodness-of-fit, while its predictive advantage is more limited.