Interval partial least-squares regression (iPLS): A comparative chemometric study with an example from near-infrared spectroscopy

A new graphically oriented local modeling procedure called interval partial least-squares (iPLS) is presented for use on spectral data. The iPLS metho d is compared to full-spectrum partial least-squares and the variable selec tion methods principal variables (PV), forward stepwise selection (FSS), an d recursively weighted regression (RWR). The methods are tested on a near-i nfrared (NIR) spectral data set recorded on 60 beer samples correlated to o riginal extract concentration. The error of the full-spectrum correlation m odel between NIR and original extract concentration was reduced by a factor of 4 with the use of iPLS (r = 0.998, and root mean square error of predic tion equal to 0.17% plate), and the graphic output contributed to the inter pretation of the chemical system under observation. The other methods teste d gave a comparable reduction in the prediction error but suffered from the interpretation advantage of the graphic interface. The intervals chosen by iPLS cover both the variables found by FSS and all possible combinations a s well as the variables found by PV and RWR, and iPLS is still able to util ize the first-order advantage.