Algorithm selection and instance space analysis for curriculum-based course timetabling
Arnaud De Coster,Nysret Musliu,Andrea Schaerf,Johannes Schoisswohl,Kate Smith-Miles
DOI: https://doi.org/10.1007/s10951-021-00701-x
2021-09-10
Journal of Scheduling
Abstract:We propose an algorithm selection approach and an instance space analysis for the well-known curriculum-based course timetabling problem (CB-CTT), which is an important problem for its application in higher education. Several state of the art algorithms exist, including both exact and metaheuristic methods. Results of these algorithms on existing instances in the literature show that there is no single algorithm outperforming the others. Therefore, a deep analysis of the strengths and weaknesses of these algorithms, depending on the instance, is an important research question. In this work, a detailed analysis of the instance space for CB-CTT is performed, charting the regions where these algorithms perform best. We further investigate the application of machine learning methods to automated algorithm selection for CB-CTT, strengthening the insights gained through the instance space analysis. For our research, we contribute new real-life instances and extend the generation of synthetic instances to better correspond to these new instances. Finally, this work shows how instance space analysis and the application of algorithm selection complement each other, underlining the value of both approaches in understanding algorithm performance.
engineering, manufacturing,operations research & management science