[Paper Review] Second International Nurse Rostering Competition (INRC-II) --- Problem Description and Rules ---

Sara Ceschia, Nguyen Dang Thi Thanh+3 more|arXiv (Cornell University)|Jan 17, 2015

Scheduling and Timetabling Solutions11 references19 citations

TL;DR

This paper introduces the Second International Nurse Rostering Competition (INRC-II), a multi-stage nurse scheduling challenge where solvers must generate weekly rosters over a fixed 4- or 8-week horizon, incorporating cumulative constraints via inter-week history. The competition uses a command-line simulation and validation stack to evaluate solutions based on hard and soft constraints, including shift patterns, skill coverage, and consecutive work/off-day limits, with final evaluation performed across the entire planning period.

ABSTRACT

In this paper, we provide all information to participate to the Second International Nurse Rostering Competition (INRC-II). First, we describe the problem formulation, which, differently from INRC-I, is a multi-stage procedure. Second, we illustrate all the necessary infrastructure do be used together with the participant's solver, including the testbed, the file formats, and the validation/simulation tools. Finally, we state the rules of the competition. All update-to-date information about the competition is available at http://mobiz.vives.be/inrc2/.

Motivation & Objective

• To design a scalable, multi-stage nurse rostering competition that models real-world planning constraints over fixed horizons.
• To define a standardized problem formulation with scenario, weekly data, and inter-week history inputs to ensure consistent solver evaluation.
• To provide a validated simulation and testing infrastructure to assess solver performance on cumulative constraints across multiple planning weeks.
• To encourage research in large-scale nurse scheduling with dynamic, history-aware constraints and practical healthcare management requirements.
• To establish a benchmark for evaluating advanced optimization techniques in real-world scheduling scenarios with complex soft constraints.

Proposed method

• The problem is structured as a sequence of weekly scheduling tasks, where each solver must generate a roster for one week based on scenario, week-specific data, and prior history.
• History data includes border states (e.g., last shift worked) and global counters (e.g., total night shifts), which are carried forward and used to evaluate cumulative constraints.
• The competition uses a command-line simulation tool that invokes the solver iteratively per week, updates history, and validates the full solution at the end.
• A validator checks the concatenated solution across all weeks, evaluating all constraints—including those based on cumulative counters—against global thresholds.
• File formats for scenario, week data, and history are standardized and documented in Appendix A for consistent input/output handling.
• Constraint evaluation is performed both at the start and end of each stage, with special handling for consecutive work days, days off, shift successions, and skill coverage.

Experimental results

Research questions

• RQ1 How do history-dependent constraints affect the performance and design of nurse rostering solvers in multi-stage scheduling problems?
• RQ2 What is the impact of cumulative constraints (e.g., total night shifts, consecutive work days) on solution quality and solver complexity?
• RQ3 How effectively can a modular, stage-wise solver handle inter-week dependencies while maintaining global feasibility?
• RQ4 To what extent can a standardized simulation and validation stack ensure fair and reproducible evaluation across diverse solver approaches?
• RQ5 How do different constraint types (e.g., forbidden shift successions, minimum/maximum consecutive days) influence the difficulty and structure of the scheduling problem?

Key findings

• The INRC-II problem formulation introduces a multi-stage approach with inter-week history, enabling modeling of long-term scheduling constraints such as cumulative shift counts and consecutive work patterns.
• The competition infrastructure includes a simulation tool that processes solvers week-by-week, maintaining and updating history, and a final validator that evaluates the complete solution across all weeks.
• Constraints such as minimum and maximum consecutive work days and days off are evaluated both at the start (using history) and end (using full week data), ensuring consistency across stages.
• Forbidden shift type successions are checked at the boundary between weeks, with violations detected when a shift sequence from the previous week violates the allowed transitions.
• The final evaluation aggregates all constraints—including global counters—across the entire planning horizon, ensuring that cumulative limits are enforced only at the end.
• The competition provides a standardized input/output format and file structure, enabling reproducible testing and fair comparison across diverse solver implementations.