Setting the mutation rate: scope and limitations of the 1/ L heuristic
Gabriela V. Ochoa · 2002
An important decision to make when designing a GA is how to set the evolutionary parameters. Among these parameters, the mutation rate has been acknowledged as the most sensitive one. All approaches so far for a near-optimal setting of the mutation rate have intrinsic limitations. A promising guideline is, however, the heuristic suggesting pm =1/L where L is the string length. This paper is a first attempt to explore the scope and limitations of this heuristic on GAs with bit-string representation. Specifically, we select two real-world domains as test problems and explore (i) whether optimal mutation rates change with time; and (ii) the interactions between the mutation rate and other evolutionary parameters. Results suggest that a constant mutation rate of 1/L is useful for a GA with a controlled ‘moderate’ selection pressure. It should be, however, revised for a weak or extremely strong selection pressure, and for a small population size. 1