Robust personnel rostering: how accurate should absenteeism predictions be?

TL;DR

The paper addresses the robustness of personnel rosters under absenteeism by embedding reserve shifts into a predict-then-optimize framework. It introduces a simulated ML methodology that evaluates roster robustness at a predetermined prediction performance, characterized by $\alpha$ (TPR), $\beta$ (TNR), and $\text{rFPR}$, without training models. Through a computational study on nurse rostering problems, it demonstrates that ML-informed rosters can outperform non-data-driven policies under reasonable performance levels, especially when skills are interchangeable; with hierarchical skills, achieving gains requires higher accuracy and skill-aware reservation planning. The approach provides practical guidance on minimum predictive performance needed to justify data-driven rostering and highlights directions for future research, including non-data-driven robust policies and application to other scheduling contexts.

Abstract

Disruptions to personnel rosters caused by absenteeism often necessitate last-minute adjustments to the employees' working hours. A common strategy to mitigate the impact of such changes is to assign employees to reserve shifts: special on-call duties during which an employee can be called in to cover for an absent employee. To maximize roster robustness, we assume a predict-then-optimize approach that uses absence predictions from a machine learning model to schedule an adequate number of reserve shifts. In this paper we propose a methodology to evaluate the robustness of rosters generated by the predict-then-optimize approach, assuming the machine learning model will make predictions at a predetermined prediction performance level. Instead of training and testing machine learning models, our methodology simulates the predictions based on a characterization of model performance. We show how this methodology can be applied to identify the minimum performance level needed for the model to outperform simple non-data-driven robust rostering policies. In a computational study on a nurse rostering problem, we demonstrate how the predict-then-optimize approach outperforms non-data-driven policies under reasonable performance requirements, particularly when employees possess interchangeable skills.

Figures (8)

• Figure 1: Confusion matrix standard notation.
• Figure 2: Schematic overview of how a simulated binary classifier determines the number of reserve shifts $c^*_d$ for each day $d \in D$ for a given $\alpha$ and $\beta$. Steps located within the red dashed rectangle form the fundamental components of the simulated ML methodology.
• Figure 3: Rostering costs and number of scheduled reserve shifts at various performance levels of the prediction model for the problem instance with uniform skills.
• Figure 4: Rerostering costs and the number of changes made during rerostering at various performance levels of the prediction model for the problem instance with uniform skills.
• Figure 5: Rerostering cost of the ML-informed approach compared to the rerostering cost of the baseline policies for the problem instance with uniform skills.