I am using extreme learning machine (ELM) for regression.

However, the accuracy of my ELM code is not that great.

I tried to approximate the input-output relationship of a sample dataset using ELM. In the testing phase the correlation coefficient is just 0.82. With the feed-forward neural network, I get a correlation coefficient close to 0.99.

I have used three variants of ELM: basic ELM, Kernel ELM, and regularized ELM.

I am looking for answers from the expert of this area.