Modeling Spectral Data Based on Mutual Information and Kernel Extreme Learning Machines

摘要

Effective modeling based on the high dimensional data needs feature selection and fast learning speed. Aim at this problem, a novel modeling approach based on mutual information and extreme learning machines is proposed in this paper. Simple mutual information based feature selection method is integrated with the fast learning kernel based extreme learning machines to obtain better modeling performance. In the method, optimal number of the features and learning parameters of models are selected simultaneously. The simulation results based on the near-infrared spectrum show that the proposed approach has better prediction performance and fast leaning speed.