We investigated four greedy algorithms for selecting the most informative features for solving the problem of multiclass classification. The algorithms have been experimentally tested on images from the Kylberg Texture Dataset [1]. The formation of features was carried out using the MaZda software, which allows calculating the texture characteristics of the image. With the help of the algorithm of greedy forward selection, it was possible to reduce the dimension of the feature space from 298 to 141 features, and the proportion of correctly classified objects increased from 85% to 96%.

1.
G.
Kylberg
.
The Kylberg Texture Dataset v. 1.0, Centre for Image Analysis, Swedish University of Agricultural Sciences and Uppsala University
,
External report (Blue series)
35
(
2011
) http://www.cb.uu.se
2.
M.
Szczypinski
,
M.
Strzelecki
, and
A.
Materka
, “
MaZda – a Software for Texture Analysis
”, in
International Symposium on Information Technology Convergence
,
2007
, p.
6
.
3.
M.
Ferńandez-Delgado
,
E.
Cernadas
,
S.
Barro
, and
D.
Amorim
,
Journal of Machine Learning Research
15
,
3133
3181
(
2014
).
This content is only available via PDF.
You do not currently have access to this content.