Partial least-squares methods for spectral analyses. 2. Application to simulated and glass spectral data

摘要: Partial least-squares (PLS) methods for quantitative spectral analyses are compared with classical (CLS) and principal component regression (PCR) by using simulated data infrared spectra from bulk seven-component, silicate-based glasses. Analyses of the sets show effect pretreatment, base-line variations, calibration design, constrained mixtures on PLS PCR prediction errors model complexity. also illustrate some qualitative differences between PSL PCR. predicted concentration a set Fourier transform glasses (S-glass) that not statistically different these two individual limited numbers samples. However, both superior to CLS in case analysis S-glass where only one analyte is known samples components unknown overlap all features components. precision significantly improves when three concentrations (B/sub 2/O/sub 3/, P/sub 5/, OH) used calibration. In this latter case, predictions unchanged, andmore » although they each still yield lower standard error than method, there no longer strong statistical evidence or outside experimental B/sub 3/ component. The ability provide chemically useful estimates pure-component demonstrated.« less