An improved ensemble model for the quantitative analysis of infrared spectra

Abstract Spectroscopy is widely applied in fast and non-destructive quantitative analysis fields. However, the translation of raw spectra into a final analysis remains a complex challenge. This paper presents a derivative spectra modeling method based on a singular perturbation technique, termed ensemble derivative spectra partial least squares (EDSPLS), for the quantitative analysis of infrared spectra. EDSPLS focuses on obtaining a final ensemble model by making full use of derivative spectra information. The algorithm combines the advantages of derivative spectra, interval PLS and fusion modeling. The process of EDSPLS includes two steps of stacked PLS. First, the inner-stack step is performed on the candidate sub-interval spectra of the zero-order, first-order and second-order derivative spectra space. Second, in the outer-stack step, three stacked models generated from the different derivative spectra space are integrated by certain weighting rules to obtain the final model. Experimental results on two public infrared spectra datasets demonstrate that the proposed EDSPLS provides superior predictive power and outperforms the conventional PLS method. Compared with the single model, the new ensemble model can provide more robust prediction results and can be considered to be an alternative choice for quantitative analytical applications.