Building Bayesian Networks for Problems of Risk Attributable to Natural Hazards

Abstract

Building Bayesian networks (BNs) for problems of risk attributable to natural hazards is a complex problem. Risk analysis makes use of domain experts’ knowledge, which is an essential element of input for the analysis. This knowledge is acquired from fragmented sources of expertise. The effectiveness of the analysis depends on the integration of the knowledge and data, which is of vital significance to decision makers. Setting up the flow of information for the acquired knowledge, which is from different fragmented sources, is a difficult task. Moreover, the process of dealing with the fragmented knowledge of domain experts may be equally complicated because of the large number of variables involved in the case of problems of risk attributable to natural hazards. To this end, a new approach using graph-theoretic techniques is proposed in this paper for integrating experts’ knowledge and data for setting up the flow of information for building BNs, so that it can be used in risk analysis. To demonstrate the approach, a case study on windstorm-induced damage of a roof structure is considered.