Nurse scheduling problem: Investigating the principles of operators in evolutionary algorithm for small size population

Abstract

Developing an effective nurse shifts assignment system, that considers diverse nurse preferences and fairness as well as ward coverage in practical operational scenarios, is a complex and time-consuming task. Failing to address various constraints with different levels of precedence can lead to undesirable nurse schedules. The efficiency of such a system relies heavily on the attributes of an automated scheduling approach or the proficiency of a head nurse. Therefore, this paper investigates the principles of designing artificial computing operators for a matrix representation solution in the evolutionary algorithm hybridization. Several parent selections, each with different selection intensities that prioritize elitism and dissimilarity characteristics, are reviewed. Additionally, the integration of parent selection intensities with specific fragment sizes of crossovers are studied when designing a well-performing algorithm. The evaluation criteria encompass algorithm reliability, accuracy, effectiveness, and efficiency. The study reveals that the modified Maximax and Maximin parent selection with Block-wise crossover achieved a higher quality schedule with the lowest fitness value. In conclusion, a small-sized population proves suitable for addressing the complex computational problem that consist of heavy constraints. The selection intensity should strike a balance between elitism and dissimilarity intensities when combined with a smaller fragment size of mating strategy.