Diagnostic reasoning based on a genetic algorithm operating in a Bayesian belief network

The feasibility of diagnostic reasoning in a Bayesian belief network, based on a genetic algorithm is demonstrated. The reasoning process described here is an example of approximate reasoning. Since exact abduction in a network modelling the ''classical diagnostic problem'' is NP-hard, inexact or approximate reasoning attracts much attention. The results of the present study indicate that in a given context of observed symptoms, a genetically generated population of possible solutions retains much of the diagnostic power contained in the full model: the disease probabilities as occuring in this population and as calculated from the full model are strongly rank-correlated. Moreover, the disease-symptom correlations are retained in the genetically generated population. This is important, since these probabilities and correlations are dynamic quantities which depend on the context of observed symptoms. The genetic algorithm may be seen as a procedure to dynamically generate diagnostic protocols.