Abstract
We propose a method of feature extraction to improve the performance of pattern recognition technique. The extracted features are defined as polynomial expressions, which are composed of the original input information. These polynomial expressions are searched by a coevolutionary genetic programming. We introduce a new fitness function based on competition between individuals. Experiments were performed for real-world datasets with k nearest neighbor classification rule. From experimental results, we have confirmed that the proposed method could maintain the diversity of populations and improve discrimination accuracy on most datasets.
