Creating and sharing knowledge for telecommunications

A Technique for Unsupervised Feature Discretization and Selection

Ferreira, A. ; Figueiredo, M. A. T.

A Technique for Unsupervised Feature Discretization and Selection, Proc Portuguese Conf. on Pattern Recognition - RecPad, Vila Real, Portugal, Vol. , pp. - , October, 2010.

Digital Object Identifier:

Download Full text PDF ( 36 KBs)

Abstract
The technique of feature discretization (FD) has been
used with the benefits of reducing the amount of memory as
well as the training time, while improving classification accuracy.
There are two commonly used techniques for scalar
FD, named equal-interval binning and equal-frequency binning.
In this paper, we propose another unsupervised FD
technique using the Lloyd-Max algorithm along with a new
criterion for feature selection (FS) based on discretized features.
The experimental results show improvement compared
to other FS and FD approaches, on standard benchmarks.
Our method also works in unsupervised and semisupervised
problems, without any modification, since it does
not use the class labels.