Bayesian inference for fault based software reliability models given software metrics data

Abstract

We wish to predict the number of faults N and the time to next failure of a piece of software. Software metrics data are used to estimate the prior mean of N via a Poisson regression model. Given failure time data and a some well known fault based models for interfailure times, we show how to sample the relevant posterior distributions via Gibbs sampling using the package Winbugs. Our approach is illustrated with an example.