Evolutionary Feature Space Transformation Using Type-Restricted Generators

Abstract

Data preprocessing, especially in terms of feature selection and generation, is an important issue in data mining and knowledge discovery tasks. Genetic algorithms proved to work well on feature selection problems where the search space produced by the initial feature set already contains the target hypothesis. In cases where this precondition is not fulfilled, one needs to construct new features to adequately extend the search space. As a solution to this representation problem, we introduce a framework combining feature selection and type-restricted feature generation in a wrapper-based approach using a modified canonical genetic algorithm for the feature space transformation and an inductive learner for the evaluation of the constructed feature set.