EXPERIMENTAL-ANALYSIS OF LARGE BELIEF NETWORKS FOR MEDICAL DIAGNOSIS

We present an experimental analysis of two parameters that are importa nt in knowledge engineering for large belief networks. We conducted th e experiments on a network derived from the Internist-1 medical knowle dge base. In this network, a generalization of the noisy-OR gate is us ed to model causal independence for the multi-valued variables, and le ak probabilities are used to represent the nonspecified causes of inte rmediate states and findings. We study two network parameters, (1) the parameter governing the assignment of probability values to the netwo rk, and (2) the parameter denoting whether the network nodes represent variables with two or more than two values. The experimental results demonstrate that the binary simplification computes diagnoses with sim ilar accuracy to the full multivalued network. We discuss the implicat ions of these parameters, as well other network parameters, for knowle dge engineering for medical applications.