Mixed Decision Trees: Minimizing Knowledge Representation Bias in LCS

Xavier Llor๠and Stewart W. Wilson²

¹Illinois Genetic Algorithms Laboratory (IlliGAL), National Center for Supercomputing Applications, University of Illinois at Urbana-Champaign, Urbana, IL 61801.
xllora@illigal.ge.uiuc.edu

²Prediction Dynamics, Concord, MA, Department of General Engineering, University of Illinois at Urbana-Champaign, Urbana, IL 61801.
wilson@prediction-dynamics.com

Abstract. Learning classifier systems tend to inherit—a priori—a given knowledge representation language for expressing the concepts to learn. Hence, even before getting started, this choice biases what can be learned, becoming critical for some real-world applications like data mining. However, such bias may be minimized by hybridizing different knowledge representations via evolutionary mixing. This paper presents a first attempt to produce an evolutionary framework that evolves mixed decision trees of heterogeneous knowledge representations.

LNCS 3103, p. 797 ff.

Full article in PDF

lncs@springer.de
© Springer-Verlag Berlin Heidelberg 2004