Fuzzy Nearest Neighbor Partitioning Neural Network for Classification

The fuzzy nearest neighbor partitioning neural network (FNNP) is proposed to promote the capability of the neural network classifier. The fixed centroids problem restrains the further improvement of the neural network classifier. The original nearest neighbor partitioning method (NNP) have been proposed to address this problem. In the NNP, the learning method is employed to train the neural network in according with the distribution information of samples. However, the distribution information underutilization problem affects the learning method to obtains the neural network with expected mapping performance. Therefore, we propose the FNNP to overcome this problem. In the FNNP, the fuzzy logic theory is adopted to assist the learning method to comprehensively collect neglected distribution information that increases the probability to find the optimal neural network with expected mapping performance. Experiment results demonstrate that the FNNP achieves remarkable classification performance on various indictors.