Abstract:Feature selection is an important preprocessing procedure in machine
learning. The selection of feature subset is based on the evaluation
of different features. Relevancy and redundancy are two common-used
evaluation criteria for feature selection. Variable interaction is
another relationship between variables. Variables interaction also
has crucial effect on feature selection. Although there are a lot of
researches on variable interaction measure, there is a little works
in applying variable interaction measure to feature selection. In
this paper, a novel algorithm, which implements Feature Selection
with considering the Variable Interaction (FSwVI), is proposed. The
proposed method uses the result of variable interaction measure to
complement and improve the feature subset selected by a normal
feature selection method. The empirical results show that it is
necessary and effective to select features with considering
variables interaction.