A Comparison of the Performance of Different Metaheuristics on the Timetabling Problem

The main goal of this paper is to attempt an unbiased comparison of the performance of straightforward implementations of five different metaheuristics on a university course timetabling problem. In particular, the metaheuristics under consideration are Evolutionary Algorithms, Ant Colony Optimization, Iterated Local Search, Simulated Annealing, and Tabu Search. To attempt fairness, the implementations of all the algorithms use a common solution representation, and a common neighbourhood structure or local search. The results show that no metaheuristic is best on all the timetabling instances considered. Moreover, even when instances are very similar, from the point of view of the instance generator, it is not possible to predict the best metaheuristic, even if some trends appear when focusing on particular instance classes. These results underline the difficulty of finding the best metaheuristics even for very restricted classes of timetabling problem.