Traditionally, fast screening for the presence of cocaine in unknown powders is performed by means of colour tests. The major drawbacks of these tests are subjective colour evaluation depending on the operator ('50 shades of blue') and a lack of selectivity. An alternative fast screening technique is Fourier Transform InfraRed (FTIR) spectrometry. This technique provides spectra that are difficult to interpret without specialized expertise and shows a lack of sensitivity for the detection of cocaine in mixtures. To overcome these limitations, a portable FTIR spectrometer using Attenuated Total Reflectance (ATR) sampling was combined with a multivariate technique, called Support Vector Machines (SVM). Representative street drug powders (n = 482), seized during the period January 2013 to July 2015, and reference powders (n = 33) were used to build and validate a classification model (n = 515) and a quantification model (n = 378). Both models were compared with the conventional chromatographic techniques. The SVM classification model showed a high sensitivity, specificity, and efficiency (99%). The SVM quantification model determined cocaine content with a root mean squared error of prediction (RMSEP) of 6% calculated over a wide working range from 4 to 99 w%. In conclusion, the developed models resulted in a clear output (cocaine detected or cocaine not detected) and a reliable estimation of the cocaine content in a wide variety of mixtures. The ATR-FTIR technique combined with SVM is a straightforward, user-friendly, and fast approach for routine classification and quantification of cocaine in seized powders. Copyright © 2016 John Wiley & Sons, Ltd.
Keywords: ATR-FTIR; PLS; SVM; chemometrics; classification; cocaine; quantification.
Copyright © 2016 John Wiley & Sons, Ltd.