A frequency-based updating strategy in Compact Genetic Algorithm

The Compact Genetic Algorithm (cGA) is one of evolutionary algorithms. There are proofs in the literature that the cGA mimics the behavior of Simple Genetic Algorithm (sGA). The cGA has a benefit in requiring almost minimal memory to store candidate solutions. It represents a population as a probability distribution instead of storing whole candidate solutions. Although the cGA has many advantages, it has a limitation on solving some problems such as deceptive problem or so called trap function. Therefore, this paper proposes an adaptation of updating strategy in the compact genetic algorithm to help the algorithm to achieve a higher solution quality with fewer evaluations. We named the proposed technique as the frequency based compact genetic algorithm (fb-cGA). The fb-cGA employs information from the past. We count frequencies and continuity of updating probabilities for both up and down. The frequencies and continuity are used to guide an updating step size. The experiment results show that our proposed method requires fewer evaluations and achieves a higher solution quality than the cGA. It can save the number of fitness evaluations up to ninefold when compared with the cGA using tournament size of 2 on 3×10 trap problem.