Predictive safety analytics: inferring aviation accident shaping factors and causation

Abstract

This paper illustrates the development of an object-oriented Bayesian network (OOBN) to integrate the safety risks contributing to an in-flight loss-of-control aviation accident. With the creation of a probabilistic model, inferences about changes to the states of the accident shaping or causal factors can be drawn quantitatively. These predictive safety inferences derive from qualitative reasoning to conclusions based on data, assumptions, and/or premises, and enable an analyst to identify the most prominent causal factors leading to a risk factor prioritization. Such an approach facilitates a mitigation portfolio study and assessment. The model also facilitates the computation of sensitivity values based on perturbations to the estimates in the conditional probability tables. Such computations lead to identifying the most sensitive causal factors with respect to an accident probability. This approach may lead to vulnerability discovery of emerging causal factors for which mitigations do not yet exist that then informs possible future R&D efforts. To illustrate the benefits of an OOBN in a large and complex aviation accident model, the in-flight loss-of-control accident framework model is presented.