Estimating confidence interval of software reliability with adaptive testing strategy

摘要

Safety critical and mission critical systems such as flight control systems, nuclear reactor plants, and medical systems software reliability is critical. In such situations, in addition to the reliability estimate, its confidence intervals (CI) also are important in investigation of the system reliability. The reliability assessment is used to estimate the reliability and the confidence intervals but the program codes that are faulty are corrected later. An unbiased reliability estimator with lower variance is required for better predictability of the software reliability. However a single reliability estimate may be inadequate for examining whether the estimate is close to the true value. Hence its CI are also to be taken in to account to know with what confidence level the true value lies in the CI. Obviously if the CI width is smaller, the estimate can be considered as better to provide the reliability measure. Adaptive testing (AT) is considered as an efficient reliability estimator that minimizes the estimator variance. However, a tight CI is not considered as an optimization goal of AT. Thus this study investigates whether the AT strategy with minimum variance estimator can provide a tight CI. This is done by a new strategy combining AT strategy with Bayesian inference (AT-BI). This approach focuses on the utilization of prior knowledge such as prior software fault density or reliability. Here a subjective confidence of reliability assessment is used to construct the optimization goal for AT-BI and the optimal reliability assessment strategy is determined y applying the Bayesian inference theory. (33 refs.)