An adaptive genetic algorithm
Eric Kee, Sarah Airey, Walling R. Cyre · 2001
This paper presents an adaptive genetic algorithm that learns to adjust some of its parameters for rapid solution based on the current state of the population. The mapping from population states to parameter values is learned during a training phase that is followed by an execution phase during which the algorithm uses the learned mapping to solve the problem. The algorithm learns which crossover rate, mutation rate, and fitness scale factor produce the best fitness growth for the problem. Two variations of the method were tested on the DeJong problem suite, and in some cases produced a 66% improvement over a non-adaptive approach.