A MDPSO-Based Constructive ELM Approach With Adjustable Influence Value

Extreme learning machine (ELM) has become a powerful machine learning approach in real-world applications. However, researchers noticed that the assignments of the hidden nodes and the scale determination of the hidden layer can affect the performance of the ELM. In this paper, we developed a new constructive method to improve the efficiency of the model by evolving the hidden nodes with multi dimension particle swarm optimization and applying an influence value for each node to constrain the effect of each node, which prevents the model construction from becoming greedy. Experiments show that the performance of the proposed algorithm relatively declines when the candidate pool consists of a small number of nodes.