Use this URL to cite or link to this record in EThOS:
Title: A neural network based construction heurisitic for the examination timetabling problem
Author: McGreevy, Mark Andrew James
ISNI:       0000 0001 3624 4988
Awarding Body: Queen's University Belfast
Current Institution: Queen's University Belfast
Date of Award: 2008
Availability of Full Text:
Access from EThOS:
The allocation of scarce resources to a set of objects such that a feasible, high quality, (preferably optimum) schedule can be constructed, has provided computer scientists with a significant ongoing research challenge. This work investigates one such task, examination timetabling, made increasingly complex with the introduction of modular courses and expanding student populations: Resource variation between institutions often presents Wlique constraints specific to the institution, and has led to the development of tailor made methodologies rather than a generalised strategy suitable for all problems. A goal of the research community and of this work is to seek a general methodology applicable to multiple problems. Building on the well-established heuristic ordering principle, this work investigates the feasibility of using a neural network to construct high quality initial timetables. A Self Organising Kohonen Map is chosen for this task. Using the Wlsupervised learning and pattern recognition properties of the Kohonen network, examinations are classified by scheduling difficulty based on a range of factors including examination characteristics and the current state of the timetable. This multi-criteria approach is a considerable enhancement to the traditional single parameter heuristic ordering often used in constructing an initial timetable. Furthermore, it offers an adaptive approach, the ordering being reviewed are each placement in response to the changing timetabling environment. Having proven the feasibility of the neural network, the approach has been tested on a range of timetabling problems commonly used by other researchers in the field, and on a further problem set containing many specialised features typically fund in real institutions. Although a simple algorithm is used to place examinations in the timetable, the results achieved are comparable with others in the field, and the technique forms a firm foundation for further development. The neural network based methodology developed in this work is entirely new and offers substantial opportunity for further development and application in many areas of timetabling.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available