A Modified Multiple Hidden Layers Extreme Learning Machine Method And Its Application

A Multiple Hidden Layers Extreme Learning Machine(MELM)was developed from the Extreme Learning Machine(ELM).ELM is the single-hidden-layer feedforward neural network and input layer of ELM was connected to its hidden layer.The weights and the bias will be produced randomly between the input layer and the hidden layer,but the weights of hidden layer connecting to input layer,was produced from Moore-Penrose.MELM was build from ELM by adding hidden layers and parameters calculation methods of each hidden layer.In order to improve MELM's generalization ability,stability and calculation ability.We proposed two improvements.(1)In order to prevent effective data loss caused by Tikhonov Regularization(TR)method.We adopted a improved TR method to enhance MELM(ITR-MELM)by adjusting parameter since it can gain optimal filter factors and a better output weight.Therefore,several ELM extension algorithms are compared based on six benchmark data.It can contribute to better result than others.(2)In order to improve the ability of processing complex data in MELM,this article suggest to use Neighbourhood Components Analysis(NCA).NCA is a method to classify and reduce the dimension of data sets.Therefore,we reduce the dimensions of high-dimensional data through NCA and then input results to MELM.Compared with other ELM algorithms,this method can obtain a higher test accuracy when the number of hidden layer nodes is small.