Qualitative Analysis Using Near - Infrared Spectroscopy A Comparison of Discriminant 6 Methods in Dissolution Testing

The bootstrap error -adjusted s ingle-sample technique IBESTl is shown to per form bet ter than the Mahalanobis d is tance met ric in qualitative near-M analysis. The BEST algorithm is designed for high-speed parallel processing supercomputers, but is also shown to operate efficiently on sing le processors. Using hypothet ica l multivariate data, the bias and RSD of the BEST and Mahalanobis metr ics were compared as a function of the number of dimensions in hyperspace and the number of training samples in the calibration set. Full near-M spectra of intact pharmaceutical tablets were analyzed successfully using the BEST to predict the tablet dissu lution rate. The BEST metric is calculated more rapidly and more precisely than the Mahalanobis metric when full near4R spectra are employed in qualitat ive ana lys is .