Improving quantitative performance of principal component regression by combining multiple scores from independent spectral ranges

Abstract

When a spectral range or multiple ranges are determined for PLS or PCR, scores generated from the determined spectral ranges are used for regression to build a calibration model. However, the possible improvement in quantitative performance when independent scores from separate spectral ranges are combined together, rather than use of a series of scores in a given spectral range, has not been investigated. Possibly the combination of more descriptive scores from different spectral ranges could improve the correlation between the variation of spectral feature and the corresponding concentration variation. The proposed strategy is named as Independent Score Combination PCR (ISC-PCR). To test the performance of ISC-PCR, we employed two different sample sets (naphtha and ethanol solution) collected by NIR spectroscopy