Informs Annual Meeting 2017

MC43

INFORMS Houston – 2017

MC40

day in the future. Due to the different urgency levels of the requests, ranging from hours to weeks and months, the MRI planners must ensure that there is enough available time at any given day to schedule urgent patients, while maintaining a high utilization. We develop an optimization model that schedules non-urgent patients and facilitates the scheduling of urgent patients. The model is run in a rolling horizon framework, and is tested on real data from a Norwegian hospital. We show how the model can be used as decision support in the scheduling process.

352E Daniel H. Wagner Prize Competition II Invited: Daniel H. Wagner Prize Competition Invited Session Chair: Patricia Neri, SAS Institute, Inc., 104 Grandtree Ct., Cary, NC, 27519, United States, patricia.neri@sas.com 1 - Crew Decision Assist: Optimizing Crew Deployment for Freight Trains Dasaradh Mallampati, BNSF, Ft. Worth, TX, United States, dasaradh.mallampati@bnsf.com, Brian Roth, Anant Balakrishnan, Pooja Dewan, April Kuo, Juan Morales Crew costs account for a significant portion of train operating expenses, and so effective crew deployment is an important priority for railroad companies. Motivated by BNSF Railway’s desire to replace the current manual crew planning process with a systematic and efficient approach, we developed an optimization model and solution methodology, and implemented a system called Crew Decision Assist (CDA) to support crew scheduling for single-ended districts at BNSF. Our approach is novel because it incorporates many practical details in train crew scheduling and accounts for uncertainty in train schedules. CDA interfaces with existing systems to quickly generate effective crew deployment plans, and is expected to yield cost savings of several million dollars per year. 2 - Optimized Scoring Systems: Towards Trust in Machine Learning for Healthcare and Criminal Justice Cynthia Rudin, Duke University, 737 E. Franklin St, Chapel Hill, NC, 27514, United States, cynthia@cs.duke.edu, Berk Ustun Questions of trust in machine learning models are becoming increasingly important, as these models are starting to be used widely for high-stakes decisions in medicine and criminal justice. Transparency of models is a key issue affecting trust. The topic of transparency in modeling is being heavily debated in the media, and there are conflicting laws on the use of black box models between the European Union and the United States. This paper reveals that: (1) There is new technology to build transparent machine learning models that are often as accurate as black-box machine learning models. (2) These methods have had impact already in medicine and criminal justice. This work calls into question the overall need for black box models in these applications. 352F Health Care, Modeling and Optimization Contributed Session Chair: Anders Gullhav, Norwegian University of Science & Technology, Trondheim, Norway, anders.gullhav@iot.ntnu.no 1 - Impact of Budget Uncertainty in a Coordinated Vaccine Market Nicholas Morris, Rochester Institute of Technology, Rochester, NY, 14623, United States, njm2868@rit.edu, Jose Batista, Ruben Proano This study evaluates the impact on total welfare in a coordinated vaccine market when countries’ procuring budgets are uncertain. This uncertainty is represented through two approaches. A simulation is used to evaluate each procurement approach on randomly generated scenarios. Each experimental instance is solved via a mixed integer nonlinear program model that generates a procurement schedule. An example is discussed on how uncertainty in the US Federal budget for vaccine procurement could impact the global vaccine market. 2 - An Approximate Dynamic Programming Approach for Patient and Process Prioritization in a Hospital Environment Lida A. Apergi, University of Maryland, 8223 Paint Branch Drive, 2243 AV. Williams, ISR, College Park, MD, 20742, United States, lapergi@rhsmith.umd.edu, John Baras, Bruce L. Golden, Kenneth Wood This research tackles the problem of patient and process prioritization in a healthcare environment with limited resources. Considering the medical cardiology department of a large medical center, a holistic model is designed and developed towards optimizing the decisions of allocating the available resources to the patients in order to minimize patients’ waiting costs. Approximate Dynamic Programming techniques are applied to optimize the aforementioned decisions over time, as well as capture the heterogeneity of the patients. 3 - Optimizing the Appointment Scheduling at MRI Labs Anders N. Gullhav, Norwegian University of Science & Technology, Department of Industrial Ec & Tech Man, Trondheim, 7491, Norway, anders.gullhav@iot.ntnu.no, Marielle Christiansen, Anders R. Eilertsen, Bjorn Nygreen, Hanna M. Selvaag We study the appointment scheduling problem, where randomly arriving examination requests are to be allocated to an MRI machine at a specific time and MC41

MC42

360A 1:30 - 2:15 Lindo Systems, Inc/ 2:15 - 3:00 Artelys Corp Invited: Vendor Tutorial Invited Session 1 - Optimization Modeling Tools from LINDO Systems Mark A.Wiley, LINDO.Systems Inc, 1415 No. Dayton Street, Chicago, IL, 60622, United States, mwiley@lindo.com Exceptional ease of use, widest range of capabilities, and flexibility have made LINDO software the tool of choice for thousands of Operations Research professionals across nearly every industry for over 30 years. LINDO offers a full range of solvers to cover all your optimization needs. The Linear Programming solvers handle million variable/constraint problems fast and reliably. The Quadratic/SOCP/Barrier solver efficiently handles quadratically constrained problems. The Integer solver works fast and reliably with LP, QP and NLP models. The Global NLP solver finds the guaranteed global optimum of nonconvex models. The Stochastic Programming solver has a full range of capabilities for planning under uncertainty. Get all the tools you need to get up and running quickly. LINDO provides a set of versatile intuitive interfaces to suit your modeling preference. What’s Best is an add-in to Excel that you can use to quickly build spreadsheet models that managers can use and understand. LINDO has a full featured modeling language for expressing complex models clearly and concisely, and it has links to Excel and databases that make data handling easy.LINDO API is a callable library that allows you to seamlessly embed the solvers into your own applications. You can pick the best tool for the job based upon who will build the application, who will use it, and where the data reside. Technical support at LINDO is responsive and thorough - whether you have questions about the software or need some guidance on handling a particular application. Get started today. Visit our booth or www.lindo.com to get more information and pick up full capacity evaluation licenses to try them out on your toughest models. 2 - Discover Artelys Knitro, the Leading Solver for Nonlinear Optimization Richard Waltz, Artelys, 150 N.Michigan Avenue, Suite 800, Chicago, IL, 60601, United States, Richard.waltz@artelys.com Artelys Knitro is the premier solver for nonlinear optimization problems. This software demonstration will highlight the latest Knitro 10.3 developments, including a new preconditioner option, advancements in solving large nonlinear least-squares models and enhancements to the R interface. In addition, we will summarize some of new features coming in Knitro 11.0. The demo will also provide an overview of how to effectively use Knitro in a variety of environments. 360B Humanitarian and Disaster Relief Logistics Sponsored: Public Sector OR Sponsored Session Chair: Christopher Zobel, Virginia Tech, Blacksburg, VA, 24061-0235, United States, czobel@vt.edu Co-Chair: Andrew N. Arnette, University of Wyoming, Laramie, WY, 82071-2000, United States, aarnette@uwyo.edu 1 - An Optimization Model for Disaster Relief Asset Pre-positioning Andrew Arnette, University of Wyoming, aarnette@uwyo.edu, Christopher Zobel This research extends previous work on improving the pre-positioning of assets used for disaster relief. It is founded on an evidence-based analytical model that utilizes multiple factors for analytically characterizing risk. This, in turn, is used to determine the potential need for the sheltering of impacted populations in the face of multiple possible natural disasters. By explicitly considering risk, the model is able to consider overall effectiveness at the same time that it ensures equity in the resource allocations. MC43

217