A Bayesian predictive software reliability model with pseudo-failures

In a previous paper, a Bayesian software reliability model with stochastically decreasing hazard rate has been presented. Within any given failure time interval, the hazard rate is a function of both total testing time as well as number of encountered failures. In this paper, to improve the predictive performance of our previously proposed model, a pseudo-failure is inserted whenever there is a period of failure-free execution equals (1 - alpha )th percentile of the predictive distribution for time until the next failure has passed. We apply the enhanced model with pseudo-failures inserted to actual software failure data and show it gives better results under the sum of square errors criteria as compared to previous Bayesian models and other existing times between failures models.