Preprocessing and an improved MIP model for examination timetabling
Résumé
We address the examination timetabling problem proposed in the second International Timetabling Competition (ITC2007). This paper presents new
preprocessing stages and an improved mixed integer mathematical model. An exam-based conflict graph in which edges represent incompatibilities between exams is used. The preprocessing stages make use of the graph, together with hard constraints and room capacities, to reveal implicit constraints. Results show that the new preprocessing stages increase up to 106 % the number of conflict constraints an have more than doubled the size of the maximum clique. The improved mixed integer model comprises fewer constraints and is therefore more memory-efficient than the model presented after the competition. We propose valid inequalities based on cliques and a Data-dependent Dual-Feasible Function (DDFF) to enhance our model. We run the tests on the twelve instances presented in the competition and the Yeditepe instances. The problem contains several criteria, which
were also tested individually to help end users decide their respective importance. When our model and the original model are compared on the different instances,
results show that the improved model yields better results for the majority of the
cases and uses less memory.