Feature selection approaches and feature reduction are very close. However, feature selection allows selecting features among a certain objective function to be optimised without transforming the features ( Sequential Forward selection SFS, GA based feature selection, PSO based feature selection...) . Feature reduction approaches allows representing features in another space, so the features are transformed (PCA, LDA..)
can we compare this approaches together?
the feature are not in the same space, i think that the comparison are not feasible, am i wrong?
Thank you in advance
When you start a new project in machine learning, you collect your data. The question is how to use this data to construct a model? Here feature extraction or selection takes place. you start to manipulate your data to find interesting features to construct a model. After that, correlation between these extracted features should be studied. This step will reduce the features if any correlation occurs among them. This is one type of reduction. Some times the number of extracted features is big. This will make training of the system ver very slow. Here, feature reduction algorithms should be used to speed up your system
As you rightly mentioned, In case of feature selection you are not modifying the original space of the feature; yet you need to find and remove the features which are do not help in discrimination. So, in feature selection, you are trying to find the subset of features which are good enough to perform the classification/recognition. However, in case of feature reduction, you can transform the features into some other space to exploit the discrimination capability in the transformed space.
Hence, though both feature selection and feature reduction work with same objective they can't be compared as the transformed space plays vital role.
You can study various feature selection algorithms (filter, wrapper and embedded) with Matlab implementation (from the first link attached). Currently, this package is available for MATLAB only. For information on each algorithm and usage instructions, please read the documentation. PCA is an example of feature extraction (reduction) technique.
PS: The second link has many good answers, too.
http://featureselection.asu.edu/software.php
https://www.researchgate.net/post/Are_these_feature_selection_methods_the_state_of_the_art/2
"the feature are not in the same space, i think that the comparison are not feasible, am i wrong?"
If the features are not comparable, i.e. not expressed by the same unit of measure, usually they are normalized.
Search "z-score normalization" in google.
Dear Lamis,
Kindly visit the following link.
gandalf.psych.umn.edu/users/schrater/schrater.../Lec17PattRec09.pdf
good Article on feature selection:
http://machinelearning.wustl.edu/mlpapers/paper_files/GuyonE03.pdf
Also refer this:
http://www.surdeanu.info/mihai/teaching/ista555-spring15/readings/yang97comparative.pdf
Dimensionality reduction Feature selection:
https://people.cs.pitt.edu/~milos/courses/cs2750-Spring04/lectures/class20.pdf
feature selection or extraction comes from extracting the image directly, means this will be original but after that in order to make some process it has to be reduce those features to another set that smaller than the original set which is feature selection set .
Feature selection you just select a subset of the original feature set, without any manipulation of the data.... on the other hand, dimensionality reduction is typically choosing a new representation within which you can describe most but not all of the variance within your data, thereby retaining the relevant information, while reducing the amount of information necessary to represent it. There are a variety of techniques for doing this including but not limited to PCA, ICA, and Matrix Feature Factorization.
- Badges
- Science method