An evolutionary algorithm hyper-heuristic for producing feasible timetables for the curriculum based university course timetabling problem

Abstract

The curriculum based course timetabling problem is a well-researched domain for which there are known benchmark data sets. Various techniques have been applied to these benchmarks in order to identify a methodology that produces the best quality timetable for one or more of the benchmark problems. The study presented in this paper takes a different approach and aims at developing a system, namely, an evolutionary algorithm (EA) hyper-heuristic, that generalizes over a set of problems rather than only producing a feasible timetable for one or more of the problems. The results of a first attempt at implementing an EA hyper-heuristic to solve the curriculum based university course timetabling problem is presented. The EA hyper-heuristic searches a heuristic space of combinations of low-level construction heuristics for feasibility, instead of a solution space. The optimal heuristic combination evolved is used to construct a solution to the timetabling problem. The EA hyper-heuristic was tested on the benchmark set of curriculum based course timetabling problems used for the second international timetabling competition. The system evolved feasible solutions for all 14 problems. The study also revealed areas for further improvement.