Fuzzy multiregression network

A novel regression model, called fuzzy multiregression network, is presented in this paper. Our network can represent the interactive relationship among fuzzy linguistic attributes and provide an effective reasoning procedure on prediction and decision making. Association among the predictive attributes in a network element is described by a nonadditive set function, while the causality between the predictive attributes and the objective attribute is determined by a user-defined fuzzy operator(s). A GA-based algorithm is developed for learning the regression coefficients in each network element. The effectiveness and performance of the learning algorithm are investigated by series of experiments. A forward inference algorithm CMB-FRA is proposed to implement decision making and training data construction.