Toward Sensor to Reproduce Human Sense - The Electronic Nose Artificial Olfaction Technology - page 138

Multilayer perception: Wrapper evaluates feature subsets on the basis of their
predictive accuracy on the PARC algorithm. This approach combined with
sequential forward selection (SFS) to select features for a multilayer-perception
regression problem. It has an ability to learn and generalize, smaller training set
requirements, fast operation, ease of implementation and therefore most com-
monly used neural network. Experimental results showed that their feature
selection procedure could find small (5-10) feature subsets with similar or better
predictive accuracy than the complete set of 49-85 features.
Radial basis function: RBF networks are used for EEG signal classification.
Because the networks train rapidly, usually orders of magnitude faster than MLP,
while exhibiting none of its training pathologies such as paralysis or local minima
problems.
Probabilistic neural network: PNN is closely related to Parzen window pdf esti-
mator. A PNN consists of several sub-networks, each of which is a Parzen window
pdf estimator for each of the classes.
- Others
K nearest neighbors is a classical classification technique widely used in pattern
reorganization problems, to determine item class. This method investigates its
neighborhood class. K-NN is a type of instance-based learning, or lazy learning
where the function is only approximated locally and all computation is deferred
until classification.
Support vector machines have very good solid foundation in statistical learning
theory, and guarantees to find the optimal decision function for a set of training
data, given a set of parameters determining the operation of SVM.
Adaptive resonance theory networks were designed to address the stability-plas-
ticity dilemma, are capable of real-time learning and classification have been
applied with some success to EN data.
Genetic algorithm, conversely, is inspired by the process of natural selection and
performs a global random search on a population of solutions.
FUZZY: The theory of fuzzy logic attempts to enable machines to deal with
imprecise language used by humans in order to describe data that may not be exact
or crisp. Fuzzy-based computational algorithms are attractive to researchers in
machine olfaction due to several sources of fuzziness can be identified in the
recognition of olfactory signals such as noisy data, imprecise measurement and
odorant sample sets which overlap in the feature space.
Discriminate function analysis (DFA), in a similar manner to PCA, also
transforms data using linear discriminate functions (LDFs).
• Clustering:
Clustering is an unsupervised learning process, search to discover spatial rela-
tionships or similarities among data samples, which may be hard to discern in
high-dimensional feature space. The process of clustering involves three basic
steps: (1) defining a dissimilarity measure between examples, typically the