Adaptive selection of heuristics for improving exam timetables

This paper presents a hyper-heuristic approach which hybridises lowlevel heuristic moves to improve timetables. Exams which cause a soft-constraint violation in the timetable are ordered and rescheduled to produce a better timetable. It is observed that both the order in which exams are rescheduled and the heuristic moves used to reschedule the exams and improve the timetable affect the quality of the solution produced. After testing different combinations in a hybrid hyper-heuristic approach, the Kempe chain move heuristic and time-slot swapping heuristic proved to be the best heuristic moves to use in a hybridisation. Similarly, it was shown that ordering the exams using Saturation Degree and breaking any ties using Largest Weighted Degree produce the best results. Based on these observations, an iterative hybrid approach is developed to adaptively hybridise the Kempe chain move and timeslot swapping heuristics in two stages. In the first stage, random heuristic sequences are generated and automatically analysed. The heuristics repeated in the best sequences are fixed while the rest are kept empty. In the second stage, sequences are generated by randomly assigning heuristics to the empty positions in an attempt to find the best heuristic sequence. Finally, the generated sequences are applied to the problem. The approach is tested on the Toronto benchmark and the exam timetabling track of the second International Timetabling Competition, to evaluate its generality. The hyper-heuristic with low-level improvement heuristics approach was found to generalise well over the two different datasets and performed comparably to the state of the art approaches.