Two-Stage Multi-Neighborhood Simulated Annealing For Uncapacitated Examination Timetabling

We propose a Simulated Annealing approach for the Examination Timetabling problem, in the classical uncapacitated formulation of Carter et al. (1996). Our solver is based on a novel combination of many neighborhoods and a principled tuning procedure performed on artificial training instances. The experimental results on realworld benchmarks show that our solver is able to improve upon all state-of-the-art search methods from the literature on most instances, running time being equal. In addition, we performed an ablation analysis, so as to identify the most important neighborhoods. Finally, we propose a novel dataset obtained by translating realworld instances for other examination timetabling formulations. Instances and solutions, along with our source code, are available on the web for inspection and future comparison.