Abstract

Niching methods extend genetic algorithms to domains that require the location of multiple solutions. This study examines and compares four niching methods --- sharing, crowding, sequential niching, and parallel hillclimbing. It focuses on the differences between parallel and sequential niching. The niching methods undergo rigorous testing on optimization and classification problems of increasing difficulty. A niching-based technique is introduced that extends genetic algorithms to classification problems. 1 INTRODUCTION Niching methods (Mahfoud, 1995) promote the formation and maintenance of stable subpopulations in genetic algorithms (GAs), allowing GAs to extend their problem-solving power to complex domains. This study examines four niching methods and compares their performances on problems in both classification and multimodal function optimization. The problems cover a wide range of difficulty levels. Classification problems are solved via a new niching-based technique. Paral...

Citations