An Enhanced Extreme Learning Machine for Supervised Learning Tasks

ABSTRACT Extreme learning machine (ELM), which is a fast and well‐known neural network algorithm, has been widely used in various disciplines. Although the performance of ELM is remarkable, the correlated and irrelevant features affect its generalization and stability performance. To address this issue, alternative approaches have been proposed, which discuss principal components regression (PCR) and ridge regression (RR) estimators with ELM. In this study, by combining PCR‐ELM and RR‐ELM, a new approach based on the r‐k class estimator is introduced to combine the advantages of both algorithms. Then, methods are given for the selection of the tuning parameters. An experimental study has been conducted using different real datasets. The performance comparisons in terms of root mean square error criterion have been done which show that the generalization and stability superiority of the proposed algorithm over the ELM, PCR‐ELM, RR‐ELM, and PCA‐ELM algorithms for the selected tuning parameter.