Abductive reasoning by constructing probabilistic deduction graphs for solving the diagnosis problem

An algorithm is proposed for finding optimal solutions of the diagnosis problem by using deduction graphs (DG) to accomplish abductions of multiple causes and multiple symptoms. The relationship among causes, symptoms, and possible intermediaries is represented by a causal network. The algorithm accomplishes the abduction by constructing a deduction graph DG(C,S) from the cause set C to the symptom set S representing the subnetwork such that the product of the prior probability, P(C), of C and the conditional probability, P(S/C), of DG(C,S) is maximized. An optimal solution is achieved by solving a 0/1 linear integer programming problem. Based on some assumptions, the algorithm can deal with a causal network involving various mutually independent deduction graphs.