Learning Classifier System Improvement Based on Probability Driven and Neural Network Driven Approaches

Abstract

Rule-based systems like Learning Classifier System are widely used in areas where data mining, data classification and pattern recognition tasks are essential. It is often difficult to address the knowledge base of these complex classifier systems, which is usually a set of classifiers. Therefore we use evolutionary processes like genetic algorithms to develop their knowledge base. We provide modified Learning Classifier System enriched by probability model to help build an appropriate knowledge base more effectively. We included a neural network into the action selection process and therefore action can be determined accordingly with a probability model. We provide simulation results which demonstrate efficiency of learning processes to compare these approaches.