Using principal component analysis to find the best calibration settings for simultaneous spectroscopic determination of several gasoline properties

A set of 160 gasoline samples was collected from commercial stations in five Brazilian states and analyzed by ASTM methods for 13 properties. Principal component analysis (PCA) was employed to investigate the effect of infrared spectral region (near or middle), calibration algorithm (principal component regression, partial least squares or multiple linear regression) and pre-processing procedure (derivative, smoothing and variable selection) in the resulting root-mean-square error of prediction (RMSEP). The PCA score plots revealed that all properties can be satisfactorily predicted by multiple linear regression in the 1600–2500 nm region, with variables selected by a genetic algorithm, using any pre-processing technique.