An Improved Pruning Algorithm for ELM Based on the PCA
Licheng Cui, Huawei Zhai, Benchao Wang · 2017
ELM is an efficient algorithm for single-hidden feed-forward networks, but, with the increase of data, it is weak in some points, such as computational complexity and network structure, so by introducing principal components analysis, a new improved ELM algorithm is proposed, PCA-ELM, it modifies the method of sensitivity analysis of hidden neurons, and reduces the size of training data, and gives a uniform standard to judge the sensitivity of each hidden neuron, in order to prune the hidden neurons with low sensitivity. Contrast test results show the PCA-ELM is better and acceptable in training time and error.