Analysis of Different Reinsertion Strategies in Steady State Genetic Algorithm

Márk Domonkos, Majd Koshakji, Ali Youssef, Issa Kalloumah, Modar Alshamali, János Botzheim

Abstract: Usually, it is uncommon to think about the reinsertion of generated offspring in evolutionary computing as a property of the algorithm that can be changed and optimized for a problem. This way reinsertion strategies are mostly overlooked properties of such algorithms. In this paper, we would like to introduce some novel versions and analyze their effectiveness. For this, a test was conducted where ten altered reinsertion strategies (including three commonly used strategies as reference and seven new strategies we came up with) were tested with a Steady State Genetic Algorithm. We tested the strategies with two kinds of population and on five different (three continuous and two discrete) benchmark functions and problems. The results show that there are strategies that work particularly badly, while others work with similar effectiveness. Also, it turns out that the number of the individuals and the nature of the problem can change the relative effectiveness of the strategies previously categorized in the effective group.