Validating Goal Models via Bayesian Networks

Goal models are an example of requirement modeling language that has been applied to support the runtime monitoring and diagnosis of software systems and to steer self-adaptive systems. When creating a goal model, requirement engineers make assumptions concerning how the goals relate to each other and when they should be considered as satisfied. In dynamic environments, however, the assumptions made in the model may be (or become) invalid. This may result in a system that does not satisfy the stakeholders' needs and, when the model is used in adaptive systems, ineffective reconfigurations. Only few and preliminary works address the automated validation of goal or requirement models. In this paper we propose the use of probabilistic models (Bayesian Networks) to determine the validity of the assumptions underlying a goal model. We employ empirical data and probabilistic inference to automatically determine a quantitative degree of validity of goal model assumptions. We illustrate the approach on a smart traffic scenario.