Developing complete conditional probability tables from fractional data for Bayesian belief networks in engineering decision making

Bayesian belief networks (BBN) can be a very powerful technique for decision-making in construction management due to the well-established theoretical foundation and reasoning processes. A major barrier to apply BBN in construction management is the scarcity of data for setting up the networks, which necessitates the involvement of domain experts for the network structure and conditional probability tables. However, the number of probabilities required from the domain expert increases dramatically when the network becomes complex and sometimes it becomes an intractable task for a domain expert to provide the huge quantity of probabilities required in a consistent way. Therefore, this research is focused on developing the means of using fractional or incomplete data to interpolate the whole domain to facilitate and expedite the process of knowledge elicitation.; To fulfil the objective, the research included (1) investigating the method of interpolating incomplete data from one domain expert into a complete set of probabilities; (2) exploring the method of integrating incomplete data from different domain experts into a complete set of probabilities; (3) examining the possible cognitive biases of a domain expert in probability elicitation. Naive BBN were used in two parallel studies on independent knowledge domains, namely, airport development and career plans of university graduates. Domain knowledge was collected using interviews and questionnaires.; The major issues investigated included: the difference between domain experts; the inter-consistency and intra-consistency for each domain expert; the pattern of probability variation; the tendency of the domain experts' responses; and the probability distribution with the existence of a dominant factor in the network. The method of piecewise representation was recommended for developing complete conditional probability tables from fractional data for Bayesian belief networks in engineering decision making.