Using Bayesian statistics to support testing of software systems

Abstract

Testing of software systems is crucial for reliability, but is often difficult owing to the many uncertainties and practical constraints involved. The nature of the software development process often prevents straightforward application of standard statistical methods. In recent years, the current authors have developed a Bayesian statistical approach to support software testers, in particular for high reliability testing of software systems near the end of the development process. Throughout this period, they have been collaborating with an industrial partner to ensure practical relevance of the methods. An important difference when compared with existing methods to support software testing is that the judgements and current practice of the software testers are central, so the method helps testers to structure their judgements and optimize corresponding test actions, the latter process being far too complex to be done effectively without the support of such sophisticated statistical methods. In the present paper, an overview is given of this approach and of results to date. Recent progress is also discussed, together with related future research topics aimed at wide practical applicability of the approach.