Effectiveness of NSGA-II with Linearly Scheduled Pareto-Partial Dominance for Practical Many-Objecitve Nurse Scheduling

Abstract

This paper describes an application of NSGA-II as one of Multi-Objective Evolutionary Algorithms (MOEAs) to a Many-Objective Nurse Scheduling in an actual hospitals in Japan and its effectiveness. Although many techniques for the actual nurse scheduling have been poposed, they are based on the culture of work styles in Europe or in the US, and then they are not fitted for creating a nurse work schedule in Japan. The nurse scheduling problem has many objectives, twelve objectives specially in the problem shown in this paper. Such an optimization problem having many objectives is generally called a Many-Objective Optimization Problem (MaOP), and it is considered that MOEAs such as NSGA-II are not effective. Although MOEA/D and NSGA-III, which are one of MaOEA, are known as effective algorithms for MaOPs, these algorithms, for example, require an so many number of scalarization vectors or appropriate reference set, they are not always easy to apply to real world problems. The MaOEAs are also very sensitive techniques to the vectors or reference set. On the other hand, although it has been pointed out that MOEAs are not suitable for MaOP in verification reports with several benchmarks, there is no fact that MOEAs have been applied to real-world MaOPs and their effectiveness has been denied. Therefore, this paper tries to apply NSGA-II, one of MOEAs, to the practical nurse scheduling problem without omitting or reducing all the objectives, and verify its effectiveness.