Reducing Complex Attribute Interaction through Non-Algebraic Feature Construction

L.S. Shafti and E. Pérez (Spain)


Machine Learning, Genetic Algorithms, Feature Selection and Construction.


The importance of preprocessing data before looking for patterns is greatest when data representation is primitive. If lack of domain experts prevents the use of highly infor mative attributes, patterns are hard to uncover due to com plex attribute interactions. Feature construction intends to create new features that encapsulate and highlight the hid den interactions. However, its success often relies on the appropriateness of a given set of algebraic operators for ex pressing the relevant combination of attributes in the cur rent domain. When lacking prior knowledge of appropri ate operators, systems use non-algebraic feature construc tion techniques to extract features directly from training data. The paper analyzes two such systems, MFE2/GA and HINT, concluding that their different design compo nents suggest complementary functionalities. This is sup ported by an empirical system comparison using synthetic and real-world data where attribute interaction prevails.

Important Links:

Go Back