Many M/C learning algorithms perform discretization of continuous data before performing a feature selection operation. WEKA uses discretization by default.
Does anyone know which method it uses? Is it Entropy Minimization Discretization (EMD), Proportional k-Interval Discretization (PKID), Equal Width Discretization (EWD) or Equal Frequency Discretization (EFD)?
Weka has two distinct ways to discretize features: (1) the supervised approach using 'multi-interval discretization', (MDL), from Fayyad and Irani (1993), and (2) the unsupervised approach using simple binning.
I think you would have to apply a discretization filter method explicitly; I'm not aware that Weka does it by default!
@Andreu Sancho-Asensio: Is there any option in Weka to choose the discretizer?
Yes, it is in the filter menu: Explorer -> Filter -> Choose. Here you can select the desired kind of discretization (i.e., supervised or unsupervised). Alternatively, you can do it via command line and/or programatically. Here is an example http://weka.wikispaces.com/Discretizing+datasets
- Badges
- Science topic