Optimal mutation rates and selection pressure in genetic algorithms
Gabriela V. Ochoa, Inman R. Harvey, Hilary Buxton · 2000
It has been argued that optimal per-locus mutation rates in GAs are proportional to selection pressure and the reciprocal of genotype length. In this paper we suggest that the notion of error threshold, borrowed from molecular evolution, sheds new light onthis argument. We show empirically the existence of error thresholds in GAs running on a simple abstract landscape � and then investigate a real-world industrial problem, demonstrating comparable phenomena in a practical application. We study the correspondence between error thresholds and optimal mutation rates on these two problems, and explore the e ect of di erent selection pressures. Results suggest that error thresholds and optimal mutation rates are indeed correlated. Moreover, as the selection pressure increases, both error thresholds and optimal mutation rates increase. These ndings mayhaveprac tical consequences, as heuristics for measuring error thresholds in real-world applications will provide useful guidelines for setting optimal mutation rates. 1