Genetic algorithm based on independent component analysis for global optimization

We design a new Genetic Algorithm based on Independent Component Analysis for unconstrained global optimization of continuous function. We use Independent Component Analysis to linearly transform the original dimensions of the problem into new components which are independent from each other with respect to the fitness. We project the population on the independent components and obtain corresponding sub-populations. We apply genetic operators on the sub-populations to generate new sub-populations, and combine them as a new population. In other words, we use Genetic Algorithm to find the optima on the independent components, and combine the optima as the global optimum for the problem. As we actually reduce the original high-dimensional problem into subproblems of much fewer dimensions, the solution space decreases exponentially and thus the problem becomes easier for Genetic Algorithm to solve. The experiment results verified that our algorithm produced optimal or close-to-optimal solutions better than or comparable to those produced by some of other Genetic Algorithms and it required much less fitness evaluations of individuals.