Researchers have studied the nurse rostering problem (NRP) for decades, producing many efficient algorithms for different formulations of the problem. In spite of these successes, the models and methods developed seldom make it to practical implementation. We believe too much emphasis has been put on solving the problem, and that the research community needs to take a step back and reflect on which problem they are solving. When modeling the NRP, we need to temper the violation of multiple constraints, and researchers tend to formulate the objective as minimizing the weighted sum of these violations. Although a solver may view a solution as optimal, practitioners might reject it if the combination of soft constraint violations is not acceptable. In this paper, we discuss some of the major drawbacks of this modeling approach and suggest a paradigm shift that may help us overcome them. We introduce the concepts targets and acceptance thresholds and discuss how we can integrate them into nurse rostering models and solution methods. We devote some attention to setting accurate acceptance thresholds, i.e., ensuring an acceptable combination of violations without the problem becoming infeasible. As this paradigm shift involves terms that practitioners can intuitively understand, it makes the automated scheduling approaches more accessible to them. Furthermore, the resulting approach can be employed for quality control, and we argue that it can yield more reliable rostering systems in practice.
|Number of pages||15|
|Publication status||Published - 2019|
|Event||The 9th Multidisciplinary International Conference on Scheduling: Theory and Applications - Ningbo, China|
Duration: 12 Dec 2019 → 15 Dec 2019
|Conference||The 9th Multidisciplinary International Conference on Scheduling: Theory and Applications|
|Period||12/12/2019 → 15/12/2019|