194. Warm-Starting Outpatient Appointment Scheduling Using Historical Solutions
Invited abstract in session ME-3: Appointment scheduling, stream Sessions.
Monday, 15:30-17:00Room: NTNU, Realfagbygget R9
Authors (first author is the speaker)
| 1. | Sara Cambiaghi
|
| Matematica, Università di Pavia | |
| 2. | Davide Duma
|
| Dipartimento di Matematica "Felice Casorati", Università degli Studi di Pavia |
Abstract
Most Outpatient Appointment Scheduling (OAS) problems require a recurring optimization task: for each planning period, a new but structurally similar instance of the problem must be solved. Typically, near-optimal solutions for new instances are computed from scratch, without leveraging the knowledge embedded in past solutions. In this talk, we present a novel approach that applies machine learning to exploit historical scheduling solutions, accelerating the optimization process and improving the quality of future schedules.
As a case study, we examine an OAS problem for the CT-scan service in the context of Emergency Departments. In this setting, efficient resource allocation is critical due to the competing demands of three patient categories (outpatients, inpatients, and emergencies) that differ in terms of uncertainty, urgency, and needs.
We propose a stochastic programming model aimed at minimizing outpatient waiting times, inpatient and emergency completion times, and staff overtime. To solve this complex problem efficiently, we employ a non-dominated sorting genetic algorithm. Additionally, we introduce an optimal assignment-based k-Nearest Neighbor algorithm to generate first-generation solutions as a warm start, enhancing the approximation of the Pareto front.
A computational analysis based on real-world data from a regional trauma hub in Pavia, Italy, demonstrates the effectiveness of the proposed approach.
Keywords
- Patient scheduling
- Optimization algorithms
- Artificial Intelligence
Status: accepted
Back to the list of papers