Document Type

Honors Project - Open Access


Graphic illustrating the concept of roulette wheel selection used with permission of John Dalton, Newcastle University, UK.


Traditionally, academic institutions schedule courses using constraints that ensure that instructors and courses do not overlap in available rooms and time periods; students' planning needs are rarely taken into account. This problem becomes particularly acute for students in liberal arts institutions, because they have multiple graduation requirements in addition to their chosen academic program. My research builds on the University Course Timetabling Problem (UCTP) to include students' scheduling needs. This approach to the UCTP problem uses a combination of a genetic algorithm and case-based reasoning.

To improve the performance of the genetic algorithm, I use a group-based genetic algorithm to place courses into distinct rooms and a self-fertilization crossover operator to avoid adding duplicate courses to the timetable during crossover. Case-based reasoning serves as a system to store and retrieve previous solutions. If a new problem is given, instead of using a genetic algorithm to produce timetables from scratch, the system first checks if the case-base has a previous timetable that solves the problem. I generate test data using knowl- edge of class scheduling at Macalester College. Although the student constraint is harder to satisfy than the instructor constraint, my results show that the genetic algorithm improves the fitness of the population for each generation, and it returns a feasible solution, even for the most constrained benchmarks.



© Copyright is owned by author of this document