MDL-based fitness for feature construction

Authors:
Leila Shila Shafti;Eduardo Pérez
Affiliations:
Universidad Autonoma de Madrid, Madrid, Spain;Universidad Autonoma de Madrid, Madrid, Spain
Venue:
Proceedings of the 9th annual conference on Genetic and evolutionary computation
Year:
2007

Citing 4
Cited 0

C4.5: programs for machine learning

C4.5: programs for machine learning
Learning by discovering concept hierarchies

Artificial Intelligence
Understanding the Crucial Role of AttributeInteraction in Data Mining

Artificial Intelligence Review
Reducing complex attribute interaction through non-algebraic feature construction

AIAP'07 Proceedings of the 25th conference on Proceedings of the 25th IASTED International Multi-Conference: artificial intelligence and applications

Quantified Score

Hi-index	0.00

Visualization

Abstract

Primitive data representation of real-world data facilitates attribute interactions, which make information opaque to most learners. Feature Construction (FC) aims to abstract and encapsulate interactions into new features and outline them to the learner. When a GA is applied to perform FC, the goal is to generate features that facilitate more accurate learning. Then the GA's fitness function should estimate the quality of the constructed features. We propose a new fitness function based on Minimum Description Length (MDL). This fitness is incorporated in MFE2/GA to improve its accuracy. The new system is compared with other systems based on Entropy or error-rate fitness.