Graph Coloring with Adaptive Genetic Algorithms

This technical report summarizes our results on solving graph coloring problems with Genetic Algorithms (GA). After testing many diierent options we conclude that the best one is a (1+1) order-based GA using an adaptation mechanism that periodically changes the tness function, thus guiding the GA through the search space. Except from the decoder ((tness function) this GA is general, using no domain speciic knowledge. We compare this GA to a powerful traditional graph coloring technique, DSatur, on a wide range of problems with diierent size, topology and edge density. The results show that the GA is superior to DSatur on the hardest problem instances and it scales up better with the problem size. The GA exhibits a linear time complexity for one measure and indicates a polynomial time complexity for another one.