Tea and herbal infusions are the most consumed non-alcoholic beverages worldwide and possess bioactive components with multiple health benefits. They are categorized in different classes that depend on: the elaboration process, origin, and components. Commonly, analytical methods are employed to classify tea according to its a chemical composition by liquid and gas chromatography-mass spectrometry, among others. Novel methods, such as electronic noses (e-noses) effectively provide real-time and objective monitoring of odors for extended periods of time. This work aimed to classify 8 different types of tea (green, white, black, spearmint, mint, hibiscus, lemongrass, chamomile) using two feature extraction methods and two pattern recognition analyses that were compared. A total of 34 tea samples were analyzed by e-nose consisting of a sample handling system as olfactometer, seven chemo-resistive gas sensors, and a 12-bit analog-to-digital converter. Tea samples were measured 10 times to ensure repeatability, resulting in database of 340 tea measures with 2499 samples each per sensor.

Data were pre-processed using Principal Component Analysis (PCA) and Parallel Factor Analysis (PARAFAC). The information extracted was classified by Artificial Neural Network (ANN) and k-nearest neighbor (k-NN). The best architecture in ANN and distance in k-NN were demonstrated by 10 k-fold cross-validation. The classification rate was 93% in ANN and PCA, 73% in ANN and PARAFAC, 94% in k-NN and PCA, and 84% in k-NN and PARAFAC. This demonstrates that conventional PCA is better than complex PARAFAC.