An empirical analysis of the grouping genetic algorithm: the timetabling case

A grouping genetic algorithm (GGA) for the university course timetabling problem is outlined. We propose six different fitness functions, all sharing the same common goal, and look at the effects that these can have on the algorithm with respect to both solution quality and time requirements. We also propose an additional, stochastic local search operator and discover that this too can have large positive and negative effects on the runs. As a byproduct of these studies, we introduce a method for measuring population diversity with the GGA model and note that diversity seems to have huge consequences on the cost implications of the algorithm. We also witness that the algorithm can behave quite differently with varying sized instances, introducing scaling-up issues that could, quite possibly, apply to grouping genetic algorithms as a whole.