Discretization method of continuous variables in Bayesian network parameter learning
Yang Feng · Systems engineering and electronics · 2012
First,based on the influence of the discretization scheme on the inference function,the conditional information entropy is proposed to evaluate the discretization scheme;second,in order to reduce the complexity of problem solving,the whole network is proposed to divide into many minimum simple sub-networks and every one is discretized respectively;third,based on the similarity between discretization and path planning,an ant colony algorithm is used to problem solving.Experimental results show that the proposed method could do well in discretization of continuous variables so as to improve the inference performance.