Information Based Feature Selection C++ Library
Feature selection is a fundamental problem in
tasks like classification, data mining, image processing, conceptual
learning. The feature selection problems refer to the task of
identifying and selecting a useful subset of features. This subset is
represent the goal from a larger set often redundant, possibly
features with different associated measurement costs and/or risks.
Algorithms of feature subset selection can be classified into three
categories based on
whether algorithms of feature selections are done independently of the
learning algorithm or not. First technique is called the filter
approach, where feature selection is performed independently of the
algorithm. When the goal is the maximization of the accuracy of a
given feature subset, we often use for the good of the subset,
function and learnig algorithm. This defines the wrapper approach.
The last type of selection approaches is the embedded approache, where
only classification algorythms with different types of heuristic
organization are used to search the important subset of features.
Durning the processes of desining selection algorithm we must take four
basic issues that determine the nature of search process into
- a starting point in features space
- an organization of the search
- an evaluation strategy of the selected
- a criterion for halting the
is based on the information theory and especially on probability, which
allow construct different types of feature selection algorythms.
Our main goal is to provide capability to do following task easily and
- Implement and test new ideas and variants
of feature ranking and selection algorithm.
- Generate simple statistics about
- Experiments with hybrid algorithm such as
Markov Blanket, FCBF and other.
- Compare algorithms on different datasets
(e.g. compare algorithms ADC,SUC and datasets from
UCI Repository or GDataset)
- Joint to the another C++ libraries as
part of them, but our library may be as a
independent from of one.
If you are interested in
InfoSel++, please get in contact with InfoSel++ Team.