Improving Software Testing with Causal Modeling
Norman Fenton, Martin Neil · 2018
This chapter introduces the idea of causal modeling using Bayesian networks (BNs), and shows they can be used to improve software testing strategies and better understand and predict software reliability. There have been many non-causal models for software quality and resource prediction, mainly in the form of regression and correlation. However, in general these models are typically data-driven statistical models. The chapter explains the basics of Bayesian reasoning and BNs. The necessary mechanism is driven by Bayes' theorem, which provides with a rational means of updating belief in some unknown hypothesis in the light of new or additional evidence. The chapter describes BNs that have been effectively used to support software testing and defects prediction. It looks at the special problems of predicting software reliability. The chapter also looks at typical approaches to risk assessment used by software project managers and demonstrate how the BN approach can provide far more logic and insight.