The estimated concentrations of chromophores (, , and ), the scattering magnitude and the scattering power , the oxygenation fraction , the estimated EMSC coefficients (, , , ), and the calculated PCA scores of the EMSC preprocessed spectra were modeled by a polynomial function of the applied contact pressure : Display Formula
(7)where is the fit error. The calculated polynomial coefficients (without the ) were used to form the feature vectors, employed by the subsequent quadratic discriminant analysis (QDA)18—based classification into three classes corresponding to the three selected measurement sites. The optimal feature subsets for the two modeling approaches were selected by the sequential forward-floating selection algorithm (SFFS),19 minimizing the average classification error rate of the training set. The classification sensitivity and specificity were estimated by the independent test set. It should be stressed that only the relative spectral changes were used in this study. Therefore, the polynomial coefficients , which were not affected by the applied contact pressure, were not used for the classification.