Sensor Circuits - The Electronic Nose Artificial Olfaction Technology - page 198

further efforts. An unsupervised data reduction method which is identified as one
of the most successful techniques is illustrated as PCA.
To describe variation of a multivariate data set in terms of a set of uncorrelated
variables is the basic principle of the PCA data analysis methods. In PCA each of
the uncorrelated data variables is a particular linear combination of the original
variables. Simply said in another way, the original data matrix is projected from a
high-dimensional space into a less-dimensional space, specifically a plane or a
three-dimensional space. In the process stage the original data set is reduced in
dimension or compressed which results in little loss of information.
A basis of a space which is represented by training vectors is calculated by the
PCA and this basis vectors are actually eigenvectors. Eigenvectors are computed
by PCA in the direction of the largest variance of the training vectors. These basis
vectors, actually eigenvectors, computed by PCA are in the direction of the largest
variance of the training vectors. Consideration of quantitative measurement of how
much a component represents the data represents eigenvalues. As the eigenvalues
of a component are higher, it represents more of the data. For simplest and robust
ways of dimensionality reduction PCA is the best solution [ 9 ].
PCA is generally related with identifying correlations in the data (Correlation
measures the simultaneous change in the values of two or more variables). There
are various models available for describing the behavioral nature of a simultaneous
change in values, such as linear, exponential, periodic, and more. In PCA the
correlation used is linear.
When obtained from procedures on a number of pragmatic variables and
development of smaller number of artificial variables identified as principal
components a suitable analysis technique is PCA. The principal components may
then be used as analyst or standard variables in subsequent analyses. Principal
component analysis is a variable reduction method. It is useful when collecting
data on a number of variables (possibly a large number of variables), and con-
sidering that there are few redundancies in those variables.
The main functions of PCA are prediction, redundancy removal, feature
extraction, data compression, etc. Applications such as signal processing, image
processing, system and control theory, communication, etc., have linear models
and appropriately implement a traditional technique PCA.
There is an option of a reliable representation of an e-nose data set in a sub-
space of reduced dimensionality produced with the fact that the chemical sensors
always exhibit a certain degree of correlation among them. The principal com-
ponent analysis consists in finding an orthogonal basis where the correlation
among sensors disappears.
Artificial Neural Network
The main characteristic of an Artificial Neural Network (ANN) model is
whether it needs direction in learning or not. Based on the way they learn, all
artificial neural networks can be divided into two learning categories: supervised
and unsupervised. In supervised learning, a preferred output result for each input