Research On Extreme Learning Machine Algorithm And Application

In recent years,the extreme learning machine(ELM)algorithm has been widely used,and many scholars pay attention to it.The extreme learning machine algorithm is a new single hidden layer feedforward neural network learning algorithm.Its learning speed is significantly better than other neural network learning algorithms,and it shows better performance in generalization.Compared with ELM,Extreme learning machine with kernel(KELM)solves the problem of random initialization of extreme learning machine(ELM),and it has faster learning speed.However,KELM has theoretical defect.When the dimensionality of kernel function mapping space is less than the number of samples,kernel function can't be introduced into ELM theoretically.Based on this condition,Extreme Learning Machine with Feature Mapping of Kernel Function(FM-KELM)is proposed in this paper.This algorithm can solve the problem that kernel function can't be introduced into ELM.In addition,this paper applies ELM algorithm to feature selection,and propose a feature selection algorithm based on ELM,the algorithm can calculate the weight of the feature,and then select features based on the weight of the feature.The main research results of this paper are as follows:This paper proves the theoretical defect of the KELM algorithm,namely,when the dimensionality of kernel function mapping space is less than the number of samples,kernel function can't be introduced into ELM.Furthermore,we theoretically prove when the regularized parameter C is small,the solution of introduced kernel function is just an approximate solution.Therefore,this paper proves a novel FM-KELM algorithm.The proposed algorithm replaces the former feature mapping between input layer and hidden layer with the newly feature mapping of kernel function.And then the problem that kernel function can't be introduced into ELM is solved.Compared with KELM,the proposed algorithm is more robust for the parameter C,and achieves higher classification accuracyIn this paper,a novel feature selection algorithm based on extreme learning machine is proposed,the proposed algorithm determines the feature weights by analyzing the input weights and output weights.Different from feature selection algorithm based on traditional neural network,the proposed algorithm mainly considers the output weights and the feature weights are mainly determined by the output weights.Furthermore,the output weights of the proposed feature selection algorithm is globally optimal.Thus,the feature weights calculated by the proposed feature selection algorithm is more reasonable than the feature weights calculated by the feature selection algorithm based on BP neural network.Four UCI data sets are used to validate the performance of the proposed feature selection algorithm.The results of the experiments demonstrate that by selecting features,the ELM achieves better generalization performance,and the computational cost is lower.