AN ENHANCED ALGORITHM FOR LINEAR MULTIVARIATE CALIBRATION

We present a new method of linear multivariate calibration that can ge nerate better prediction results than those obtained by partial least squares (PLS). This is accomplished by incorporating the spectrum of t he desired species into the calibration procedure. The method combines the advantages of different standard methods and is therefore called hybrid linear analysis (HLA). In side-by-side tests using both simulat ed and experimental data, HLA produced lower prediction errors than PL S in all instances. We recommend HLA over PIS in situations where the spectrum of the desired species is available.