Constructive versus Improvement Heuristics: An Investigation of Examination Timetabling

Examination timetabling has been a widely studied research problem for over 30 years, with many techniques being used to produce good quality solutions. The majority of the techniques rely on producing an initial solution using a constructive heuristic, and then improving it via a local search [1,2,3], relying on the correct setting of internal parameters to allow generally acceptable results for different data sets. Unfortunately, in attempting to provide good quality results for all data sets, in many cases we may not achieve the best possible solutions. From the author’s practical experience, it has been observed that human intervention is often necessary to adapt the scheduling techniques used for the particular characteristics of each data set. The goal of this paper is to remove this human intervention where possible, and allow full automation for all problem instances. We investigate a number of benchmark problem data sets, of varying degrees of complexity, to see if there is a relationship between the difficulty of the problem and the effects of combining constructive and improvement heuristics in achieving the best possible solution under specified time constraints. In this way we can attempt to generalise the link between the problem instances and the approach taken in gaining a solution. We use the Carter benchmark datasets [4], which represent real world instances and whose characteristics in terms of measure of difficulty are well known. We use the commercial examination timetabling system, Optime [5] to generate the results. Our results show that there is a relationship between the difficulty of the problem and the time spent in each part of the solution methodology.