Classification Problem Research Based On Extreme Learning Machine

In recent years,with the rapid development of multimedia technology,digital technology,the field of machine learning has been speedy developed.As a competitive machine learning algorithm,Extreme learning machine(ELM)attracts more and more scholars' attention with the characteristics of its concise theory and easy way to implement.In the classification and regression issues,ELM have been widely used and showed good results.However,in the practical problem,because of high dimension data,noise,outliers and other issues,reduces the classification accuracy of ELM.In this paper,we study the extreme learning machine for the above problems.The main research results are as follows:(1)In order to solve the problem that the data sample is too high dimension to reduce the accuracy of the classification of ELM,we propose a dimension reduction algorithm of supervised sparse locality preserving projection(SSLPP).The local linearization neighborhood range of data is dynamically determined by the calculating neighborhood information of data,and then the global and local discriminant information of the face image data is obtained accurately.This can effectively eliminate the influence of the redundant attributes contained in the data samples to the activation function,which can effectively.avoid the ill-posed problem of the output matrix of the hidden nodes.(2)Aiming at the existing problems in dealing with noise and outliers of ELM,we propose a modified fuzzy extreme learning machine(MFELM).MFELM inherits the good learning ability and generalization ability of ELM.And compared with the existing algorithm,MFELM has better robustness in dealing with noise and outliers.