Spectral Feature Selection for Data Mining

6 Spectral Feature Selection for Data Mining

tion by removing these irrelevant and redundant features. To achieve this, a

feature evaluation criterion is used with a search strategy to identify the rel-

evant features. And a selection matrix W is used to filter the original data

set and generate a reduced data set containing only the relevant features.

1

Unlike feature selection, feature extraction achieves dimensionality reduction

by combining the original features with a weight matrix W

0

to generate a

smaller set of new features.

2

In the combination process, the irrelevant and

redundant features usually receive zero or very small coefficients, therefore ...