Estimating confidence interval of software reliability with adaptive testing strategy

Abstract

Software reliability assessment is a critical problem in safety-critical and mission-critical systems. In the reliability assessment of such a system, both an accurate reliability estimate and a tight confidence interval are required. Adaptive testing (AT) is an on-line testing framework, which dynamically selects test cases from different subdomains to achieve some optimization object. Although AT has been proved effective in minimizing reliability estimator variance, its performance on providing the corresponding confidence interval has not been investigated. In order to address this issue, an AT strategy combined with Bayesian inference (AT-BI) is proposed in this study. The novel AT-BI strategy is expected to be effective in providing both a low-variance estimator and a tight confidence interval. Experiments are set up to validate the effectiveness of the AT-BI strategy.