Determination of the number of components in mixtures using a new approach incorporating chemical information

A novel index RESO, based on the ratio of eigenvalues calculated by smoothed PCA and those calculated by ordinary PCA, was developed for determining the number of compounds in mixtures. The proposed method distinguishes itself from other methods by incorporating chemical information in its framework. Its outstanding features, such as the capability of handling situations with existence of minor components or severe collinearity in spectra and confronting heteroscedastic noise to some extent, make RESO more powerful than many other indices reported so far. The results obtained from simulated and actual data sets by the proposed method show that it can deal with more complex situations than many other methods can. Copyright © 1999 John Wiley & Sons, Ltd.