abstract = "In this paper we use genetic programming (GP) for
feature selection in binary classification tasks.
Mathematical expressions built by GP transform the
feature space in a way that the relevance of subsets of
features can be measured using a simple relevance
function. We make some modifications to the standard GP
to make it explore large subsets of features when
necessary. This is done by increasing the depth limit
at run-time and at the same time trying to avoid
bloating and overfitting by some control mechanism. We
take a filter (non-wrapper) approach to exploring the
search space. Unlike most filter methods that usually
deal with single features, we explore subsets of
features. The solution of the proposed search is a
vector of Pareto-front points. Our experiments show
that a linear search over this vector can improve the
classification performance of classifiers while
decreasing their complexity.",
notes = "GECCO-2009 A joint meeting of the eighteenth
international conference on genetic algorithms
(ICGA-2009) and the fourteenth annual genetic
programming conference (GP-2009).