Using consensus interval partial least square in near infrared spectra analysis

This paper proposes a novel consensus modeling method for regression, which optimizes the weight coefficients of member models considering both error and error correlation of member models. Thus, the optimized objective function has clear physical significance. Furthermore, the root-mean-square error of cross-validation (RMSECV) and root-mean-square error of prediction (RMSEP) of the consensus model are better than any member model. Integrating this method with interval partial least squares algorithm (iPLS), the novel consensus interval partial least squares algorithm (CPLS) is achieved. The typical near infrared spectroscopy datasets are used to validate the effectiveness of CPLS. Compared to the commonly used partial least squares (PLS), iPLS and staked interval partial least squares algorithm (SPLS), CPLS produces better prediction performance. (C) 2015 Elsevier B.V. All rights reserved.