Correspondence between causality diagram and neural networks

The problem of obtaining a correspondence between causality diagram (CD) and neural networks was studied. A method of obtaining a direct correspondence between the parameters of a causality diagram and the parameters of an associated neural network has been presented. The training capabilities of a neural network were used to determine the conditional probability matrix elements required by the causality diagram. It is shown how such a correspondence is established by obtaining a mathematical function which relates the parameters of the two models. It shows the validity of the method by deriving the parameters to be used in a causality diagram constructed to combine GIS data for assessing the risk of desertification of burned forest areas in the Northeast China.