Year: 2026
Type: article
Abstract: For the nurse scheduling problem, we propose a hybrid method that combines mathematical programming and machine learning. In the method, a mixed integer programming (MIP) problem firstly provides a shift schedule, which, however, is not always acceptable to a chief nurse because there often exists a hidden condition; this is one of the major practical difficulties in nurse scheduling. The method secondly trains a deep neural network (DNN) model with practically acceptable schedules. As a result, the DNN model automatically fixes an input schedule, which is initially the one obtained from the MIP problem, so that it is acceptable. To evaluate the capability of this method, we perform a simulation involving another MIP problem, called the virtual chief nurse problem (VCNP). The VCNP implements virtually hidden conditions, such as the incompatibility of nurses, and its solutions are used as target data for training a DNN model. Through this simulation, the DNN model learns the hidden conditions, thereby modifying schedules appropriately. This demonstrates the promise of our hybrid approach. We also conducted an experiment in collaboration with a hospital’s nursing department. In the experiment, a chief nurse repeatedly fixed the initial schedule generated by the MIP solver, and a sufficiently good schedule was obtained after a few iterations. These results suggest that the proposed method can effectively capture hidden practical constraints and improve the overall acceptability of nurse schedules.
Authors: Kayoko AOI, Atsuo SUZUKI, Shungo KOICHI
Institution: Nanzan University
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: gold
Estimated APC paid: $602
OpenAlex ID: W7204478426