AN EMPIRICAL STUDY OF EARLY STOPPING IN GENETIC PROGRAMMING

Genetic Programming (GP) is an evolutionary method based on the principles of natural genetic systems. In GP, a population of individuals are randomly generated. This population is evolved through a number of generations by applying some genetic operators such as crossover, mutation and selection. Although GP has been successful applied to many real world problems, there are very few guidelines for determining when to stop GP algorithm. Traditionally, a predefined number of generation is set and GP is stopped when it reaches to the last generation. In this article, we present an empirical study of the impact of early stopping to GP performance. We propose some early stopping criteria for GP. We tested the proposed methods on a number of symbolic regression problems. Our experiment results show that using early stopping helps to maintain the generalisation capacity of GP while significantly reducing its solutions complexity and training time.