ARTIFICIAL NEURAL NETWORKS FOR DATA MINING AND FEATURE EXTRACTION - Quantitative Medical Data Analysis Using Mathematical Tools and Statistical Techniques - page 321

CHAPTER 15
ARTIFICIAL NEURAL NETWORKS FOR DATA MINING
AND FEATURE EXTRACTION
Je Knisley a , Lloyd Lee Glenn, Karl Joplin, and Patricia Carey
The Institute for Quantitative Biology,
East Tennessee State University,
P. O. Box 70663, Johnson City, TN 37614-0663, USA
E-mail: a [e-mail]
Articial Neural Networks are models of interacting neurons that can be
used as classiers with large data sets. They can also be used for feature
extraction and for reducing the dimensionality of large data sets. Den-
dritic electrotonic models can be used to suggest more robust articial
neural network models that are amenable to data mining and feature
extraction.
1. Introduction
The very large data sets now being produced by modern scientic in-
struments often require data mining techniques which go beyond the
usual methods of statistical analysis. Among the most popular data min-
ing algorithms used to classify unknown data are the techniques of clus-
tering analysis, principal components analysis, linear discriminant analy-
sis, decision trees, support vector machines, and artificial neural
networks 1 . The majority of these techniques are applied to unclassied
data sets in order to extract features from the data that cannot be directly
observed. For example, in a clustering analysis, subgroups of the data are
classied so that members of a subgroup are relatively close to each other
while remaining relatively far away from elements in other subgroups 2 .
However, if some or all of the data is classied a priori{for example, if
a data set can be classied as coming either from an experimental group or
from a control{then support vector machines (SVM) and articial neural
networks (ANN) are often the tools of choice. In both cases, data sets for
known classications are used to train the SVM or ANN so that it can