Hybrid Variable Neighborhood Approaches to Exam Timetabling

A very important factor for the success of a local search is the neighborhood employed during the search. Variable Neighborhood Search (VNS) algorithms [5] employ more than one neighborhood and change the employment of these neighborhoods during the search. It provides a very effective way of escaping from local optima and is both versatile and successful compared to other local search techniques when applied to a range of different problem domains [5]. The aim of our research is to investigate the VNS methodology and their hybridizations with Genetic Algorithms for the university exam timetabling problem. The exam timetabling problem consists of assigning a given set of exams to a number of timeslots subject to a set of constraints. These include avoiding the schedule of exams with common students to the same time, and minimizing the students taking exams scheduled too close to each other [2,3,8]. Metaheuristics represent the state-of-the-art in the last 10 years [7]. The most commonly used neighborhood in timetabling research is the single move neighborhood that reallocates single exams to new feasible timeslots. In this work we analyzed variants of neighborhood structures in VNS and demonstrated that a biased VNS is able to provide high quality performance across a number of timetabling problem instances. Furthermore this VNS approach when hybridized with a Genetic Algorithm that intelligently selects its neighborhoods is very successful compared with the state-of-the-art approaches. Indeed, the method produces some of the best known results in the literature in terms of solution quality on the standard benchmark problems.