A comparative study of evolutionary algorithms

This article describes a comparative study of Evolutionary Algorithm with Guided mutation (EA/G) against Population-Based Incremental Learning (PBIL) and Compact Genetic Algorithm (CGA). Both PBIL and CGA are representatives of Estimation of Distribution Algorithms (EDAs), a class of algorithms which uses the global statistical information effectively to sample offspring disregarding the location information of the local optimal solutions found so far. On the other hand, EA/G uses global statistical information as well as location information to sample offspring. We implemented the algorithms to build an experimental setup upon which simulations were run. The performance of the algorithms was analyzed on numerical function optimization problems and standard genetic algorithm test problems in terms of solution quality and computational cost. We found that EA/G outperformed PBIL and CGA in attaining a good-quality solution, but both PBIL and CGA performed better than EA/G in terms of computational cost.