An Adaptive Concatenated Failure Rate Model for Software Reliability

Abstract

This article introduces a software reliability model whose concatenated failure rate function is motivated via considerations that reflect an engineer's knowledge about the stochastic nature of software failures. The model is adaptive (in a sense explained), has two parameters, and has characteristics that generalize those of existing models. A Bayesian approach for estimating the model parameters and for testing hypotheses about reliability growth is proposed. The prior distributions reflect structural considerations, and Markov chain Monte Carlo techniques are used to implement the approach.

Key Words:

• Bayes factors
• Gibbs sampling
• Markov chain Monte Carlo
• Point process
• Reliability growth.