Classification Algorithm Design Based On Modified Fuzzy Neural Network And Extreme Learning Machine

The development of neural network has promoted the promising future of pattern recognition at some extent. This thesis proposes four kinds of algorithms based on fuzzy min-max neural network and extreme learning machine. Meanwhile, the entire work focuses on improving the accuracy and efficiency of classifier system.A novel adaptive fuzzy min-max neural network classifier called MDCFMN is proposed in this paper. In contrast the former work called DCFMN, the new classifier can update the network weigh value according to the distribution of input samples. Besides, the classifier adopts genetic algorithm to finish searching the optimal combination of its parameters which saves the time consuming for training and testing.Taking the complex network architecture of all kinds of precedent fuzzy min-max neural network into consideration, combined with principle component analysis (PCA) and adaptive genetic algorithm (AGA),a integrated system called AFMN can serve as a supervised and real-time classification technique. Considering the loophole in the expansion-contraction process of FMNN and GFMN and the over-complex network architecture of FMCN, AFMN maintains the original and simple architecture of FMNN for fast learning and testing while rewriting the membership function, the expansion and contraction rules for hyperbox generation. Meanwhile, principle component analysis (PCA) is adopted to finish dataset dimensionality reduction for increasing learning efficiency thus making training faster. After training, the confidence coefficient of each hyperbox is calculated base on the distribution of samples belonging to different classes within it. During classifying dataset procedure, utilizing adaptive genetic algorithm (AGA) to complete parameter optimization for AFMN can fasten the entire procedure and eliminate the step-setting obstacle caused by traversal method.Based on the fact fuzzy min-max neural network is not fast enough during training and testing, extreme learning machine is introduced to solve this practical problem. Adopting integrated neural network to improve the classification accuracy of a single network. Meanwhile, for a single extreme learning machine, the modified ELM based on particle swarm and chaos theory is proposed to improve its performance while keeping its advantage of fast running.