A Modular Neural Networks ensembling method based on fuzzy decision-making

This paper integrates the ideas of "divide and conquer" and "put the heads together", and proposes a method of selecting sub-networks on-line based on fuzzy decision making. For a given input, this method calculates the distances between the input and the centers of sub-networks firstly, and then obtains the fuzzy member degree based on distance measure, finally realizes on-line sub-networks selection by fuzzy decision-making. The group of selected sub-networks varies with different inputs. It ensembles multi sub-networks by linear method. The weights of sub-networks are optimized by reconstructing the sample space. The simulation results suggest that this method can improve the precision and generalization ability effectively.