Learning Classifier Systems Approach for Automated Discovery of Hierarchical Censored Production Rules

Abstract

This article presents Learning Classifier Systems (LCS) approach for automated discovery of Hierarchical Censored Production Rules (HCPR). A LCS is an adaptive system that learns to perform the best action given its input. By best is generally meant the action that will receive the most reward or reinforcement from the system's environment. A classifier system consists of three main components: rule and message system, apportionment of credit system, genetic algorithm (GA). In the proposed LCS, concatenate of the Hierarchical Censored Production Rule-trees form the genotype, and therefore the GA operates on a population of HCPR-trees. More recently, LCSs have proved efficient at solving automatic classification tasks. Hierarchical Censored Production Rules is a system of knowledge representation that exhibited variable certainty as well as variable specificity and offered mechanisms for handling the trade off between the two. An appropriate chromosome representation scheme, suitable genetic operators, appropriate fitness function and also appropriate credit assignment scheme is proposed to evolve the best HCPR-trees. Experimental results are presented to demonstrate the performance of the proposed system. © Springer-Verlag Berlin Heidelberg 2010.