Information Technology Reference
In-Depth Information
Chapter 4
A Geometric Approach to Feature Ranking
Based Upon Results of Effective Decision
Boundary Feature Matrix
Claudia Diamantini, Alberto Gemelli and Domenico Potena
Abstract
This chapter presents a new method of
Feature Ranking
(FR) that
calculates the
relative weight of features
in their original domain with an algorithmic
procedure. The method supports information selection of real world features and is
useful when the number of features has costs implications. The
Feature Extraction
(FE) techniques, although accurate, provide the weights of artificial features whereas
it is important to weight the real features to have readable models. The accuracy of
the ranking is also an important aspect; the
heuristics methods
, another major family
of ranking methods based on generate-and-test procedures, are by definition approx-
imate although they produce readable models. The ranking method proposed here
combines the advantages of older methods, it has at its core a feature extraction
technique based on
Effective Decision Boundary Feature Matrix
(EDBFM), which
is extended to calculate the total
weight
of the real features through a procedure
geometrically justified. The modular design of the new method allows to include
any FE technique referable to the EDBFM model; a thorough benchmarking of the
various solutions has been conducted.
·
·
Keywords
Feature ranking
Feature weight
Effective decision boundary feature
·
matrix
Classification
4.1 Introduction
The recent developments of information technology dramatically increased the capa-
bility of gathering information. This information is described by a high number of
attributes, observations or measures, generically called
features
. On the one hand this
Search WWH ::
Custom Search