A Bayesian Model Averaging Method for Software Reliability Assessment

Due to the complexity and uncertainty of software failure behavior, the selection and application of reliability models can be challenging. To tackle the challenge of software reliability modeling and prediction, researchers have investigated abundant reliability models including combined models consisting of multiple models. This paper proposes to model and assess software reliability using the Bayesian model averaging method. The proposed modeling approach is based on Bayesian theory as well as selecting existing reliability modeling methods as candidate models. The posterior probability of a model being selected is obtained by Bayesian inference. The posterior probabilities are further used as weights to average the candidate base models. The key and difficulty of applying Bayesian model average method lies in the evaluation of weights. MCMC method is used to estimate the weights in BMA for software reliability modeling and assessment.