Principal Component Analysis (PCA) is an effective unsupervised linear method to project data
from several sensors to a two-dimensional plane. Is a linear method that has been shown to be effective
for discriminating the response of an EN to simple and complex odours [17]. The neural network
model used in this study was a multilayer perceptron (MLP) that learns by using an algorithm called
backpropagation with an adaptive learning rate [18]. Therefore, in this application a MLP (Multilayer
Perceptron) network with ten hidden and one output neurons was applied to achieve a success rate in
classification a cross-validation technique called “leave one out” of order one, was implemented to
estimate the success rate in classification. A leave-one-out estimates the performance of the network in
the classification of coffee samples. This interactive validation approach generates N evaluation
procedures (1 for each measurement). For each iteration, a different measurement is left out, while the
remaining measurement (the one not used for training) is then projected onto the neural model and
classified using the already trained network. This is repeated N times (one for each measurement) so
that the final result is the average success of entire iterative process. The data pre-processing and processing were done with algorithms written in Matlab 7.5 having added a Graphical User
Interface (GUI).