feature_selection Package

feature_selection Package

Select features by learning algorithms or simple name filtering

Feature selection subsumes methods that are used to select a number of features from a set of features that are most useful in a classification context. Thus, feature selection usually utilizes supervised learning.

Feature selection methods can be split into filter and wrapper approaches. Wrapper methods for feature selection use internally a classification algorithm (e.g. SVMs) and select the subset of features that maximize the predictive performance of this classifier on the training data. In contrast, filter methods utilize heuristics like information gain or mutual information to select features that are strongly correlated to the class of the data but only weakly interrelated.

Modules

feature_filter	Reduce filters with the help of name filters
mlpy_wrapper_nodes	Use feature selection methods implemented in MLPY
random_feature_selection	Randomly select a number of features
relief	Feature selection based on the RELIEF algorithm