An Incremental ELM Algorithm Based On Instance Selection

Incremental extension of extreme learning machines(EM-ELM,I-ELM) are proposed based on Extreme Learning Machines Classifier by static hidden nodes to dynamic hidden nodes. Recently, many studies show classifier use input weight from example can improve the test accuracy and generalization performance.Since the incremental extension of extreme learning machine doesn’t use the sample characteristics as input weights,Romero, E and Toppo, D proposed using the sample characteristics as input weight classifier(SV-SFFNS). SV-SFFNS randomly select the sample’s characteristics as input weight, and it greatly improves the test accuracy and the generalization performance of the classifier.Since SV-SFFNS select the characteristics from sample without treatment and it just simply take random way, we propose the core vector machines extreme learning machine(CVM-ELM). The CVM-ELM uses the core vector machines to select the core set of the sample, and it randomly select characteristics from core set as the hidden layer’s input weight. The core set is the informational sample. The experimental results show that it can greatly improves the test accuracy and the generalization performance of the classifier by using the core set’s characteristics as input weight.