Modelo Probabilístico Bayesiano para Simular o Conhecimento de Especialistas no Controle da Ferrugem Asiática da Soja no Estado do Paraná

Abstract

The Asian rust is the main pathology of soybean culture, what makes it the object of several expert systems. This work aimed to build a probabilistic model to estimate the need and number of fungicide applications to control soybean Asian rust in Paraná using the Bayesian network formalism and knowledge engineering. The model engineering was accomplished by interviews with experts and also by the literature review, what produced a Bayesian network built with the aid of software GeNIe 2.0, where the variables, graph structure and conditional probability table of each variable were defined, what determined the influences between them. The tests made to evaluate the model were accompanied by two interviewed experts, who approved the model through proposed test cases. The results presented showed that the developed model rigorously represent the knowledge of the expert who accompanied its development, presenting common consensus among the other interviewed experts for the first fungicide application but diverging for the extra ones.