Performance of some supervised and unsupervised multivariate techniques for grouping authentic and unauthentic Viagra and Cialis

A typical application of multivariate techniques in forensic analysis consists of discriminating between authentic and unauthentic samples of seized drugs, in addition to finding similar properties in the unauthentic samples. In this paper, the performance of several methods belonging to two different classes of multivariate techniques–supervised and unsupervised techniques–were compared. The supervised techniques (ST) are the k-Nearest Neighbor (KNN), Support Vector Machine (SVM), Probabilistic Neural Networks (PNN) and Linear Discriminant Analysis (LDA); the unsupervised techniques are the k-Means CA and the Fuzzy C-Means (FCM). The methods are applied to Infrared Spectroscopy by Fourier Transform (FTIR) from authentic and unauthentic Cialis and Viagra. The FTIR data are also transformed by Principal Components Analysis (PCA) and kernel functions aimed at improving the grouping performance. ST proved to be a more reasonable choice when the analysis is conducted on the original data, while the UT led to better results when applied to transformed data.