CLASSIFICATION RESULTS OF MASS-SPECTRA OF TOXIC COMPOUNDS BY CLASS MODELING

Classification of mass spectra of toxic compounds is based on the numerical features derived from spectral data. Computerized classification on these data provides various parameters, which determine the corresponding compound(s) belonging to a certain class of toxic compounds. Pattern recognition studies on mass spectra data are often used for principal components (PC) modeling of the autocorrelation transformed mass spectra. The significant number of principal components is determined by cross validation. With this technique, for the 119 toxic compounds targeted in ambient air, the predictive accuracy for the four classes obtained by GC/MS from training and calibration data is 76.5%.