2016 INFORMS Annual Meeting Program

WB01

INFORMS Nashville – 2016

Keynote Wednesday

5 - Home Health Care Routing And Appointment Scheduling With Stochastic Service Durations Yang Zhan, Shanghai Jiao Tong University, Shanghai, 200030, China, zhanyangjy@sjtu.edu.cn, Zizhuo Wang, Guohua Wan Motivated by the practice of home health care services, we consider an integrated routing and appointment scheduling problem with random service durations. The objective of the problem is to determine the visit route and appointment times to minimize the summation of the costs of traveling and idling of the health care team and the cost of waiting of the patients. To solve the intractable problem, we propose both exact and approximate algorithms. We conduct computational experiments to assess the performance of the proposed methods on problems of practical size.The computational results show that the methods work very well.

Davidson Ballroom C-MCC SportSource Analytics Invited: Plenary, Keynote Invited Session

Chair: James Primbs, California State University Fullerton, 925 Berenice Dr, Brea, CA, 92821, United States, japrimbs@live.com 1 - SportSource Analytics Stephen Prather, SportSource Analytics, Nashville, TN, United States, team@coachesbythenumbers.com Think back about 15 years ago about how difficult it was for anyone to access large amounts of data on virtually any subject. Now, think about how easy it is today for virtually anyone to access enormous amounts of data with the click of a few buttons. We live in an extremely data rich world. We are surrounded by information and data in all walks of life. The problem with all of this “big data” is that we are really struggling in finding ways to make it small and more importantly make it USEFUL. My talk is going to be about how four guys all working full-time jobs and without a single advanced degree in any sort of statistical analysis between them were able to become the official analytic consultant to the college football playoff selection committee. This is a story of the pursuit of being useful and understanding that data is only as good as the analysis associated with it. WB01 101A-MCC Pattern Recognition Applications in Data Mining Sponsored: Data Mining Sponsored Session Chair: Cory Stasko, Massachusetts Institute of Technology, 4 Garden Court, Apt 4, Cambridge, MA, 02138, United States, cstasko@mit.edu 1 - Auto Detection Of Tool Wear Using Sequence Alignment Technique Cheng-Bang Chen, Penn State University, 445 Waupelani Dr., Apt K18, State College, PA, 16801, United States, czc184@psu.edu, Dika Handayani, Deepak Agrawal, Juxihong Julaiti Tool wear is one common criteria used to measure the machinability of a material. Manual tool wear measurement, which is still widely done, raises an issue on how reproducibility and repeatability the measurements are. In order to reduce the variation of the measurement and speed up the process, we propose a new method using edge detection, sequence mapping, and area projection to measure the wear automatically. 2 - Mini-batch Proximal Semi-stochastic Gradient Descent In Signal Processing Jie Liu, PhD Student, Lehigh University, 14 Duh Dr Apt 324, Bethlehem, PA, 18015, United States, jild13@lehigh.edu, Jakub Konecny, Peter Richtarik, Martin Takac We propose the mini-batch proximal semi-stochastic gradient descent (mS2GD). First, we provide convergence results for mS2GD and show that it maintains a complexity of O((n+ )log(1/ )), comparable to modern stochastic gradient descent methods such as SVRG, SAG, SAGA. Second, we show that mS2GD benefits from both mini-batch speedup and the simple parallel implementation. In the numerical experiments, we first compare different algorithms on public available datasets; then, we compare mS2GD with different batch sizes to illustrate efficiency of mini-batching; last, we conduct experiments on one of the popular signal processing problems—a simple image deblurring problem. 3 - Deconstructing Va Procurement And Logistics Policy With Natural Language Processing Cory Stasko, Massachusetts Institute of Technology, 4 Garden Court, Apt 4, Cambridge, MA, 02138, United States, cstasko@mit.edu Over 120 policy documents of are involved in governing VHA procurement and logistics. This large volume of active policy makes it difficult for individuals to understand what exists, where, and how it affects them. Furthermore, the policy set includes redundancies, missing elements, and other weaknesses. This work investigates the value of natural language processing in deconstructing and mapping interrelated policy texts. We describe and organize the logical, linguistic, and substantive patterns within and between policy documents, thereby producing a dynamic map of policy evolution that highlights patterns, inter- dependencies, conflicts, ambiguities, and redundancies in the text. Wednesday, 11:00AM - 12:30PM

Wednesday, 10:00AM - 10:50AM

Keynote Wednesday

Davidson Ballroom A-MCC The Goals of Analysis are Understanding, Decisions, and Influencing Policy Invited: Keynote Invited Session Chair: Turgay Ayer, Georgia Institute of Technology, Atlanta, GA (Healthcare Analytics Chair; ayer@isye@gatech.edu 1 - The Goals Of Analysis are Understanding, Decisions, And Influencing Policy Gerald G. Brown, Naval Postgraduate School, Monterey, CA, United States, gbrown@nps.navy.mil While we are variously skilled at applying a diverse set of mathematical tools to analysis, we all share (or should share) the same goals: understand the problem at hand; advise decisions influencing that problem; and influence policy for dealing with entire classes of problems resembling the one we analyze. Sometimes, our answers are not welcomed by a client who brings us a problem, and we face significant obstacles to conveying good, convincing advice and thus contributing to good decision policy. There are a number of techniques that apply to such situations and cross all our various analysis domains. Few of these appear in textbooks or our open literature. These turn out to be vitally important for success. Davidson Ballroom B-MCC Can Prediction be Better than Cure? On Analytics in Health-Care Invited: Plenary, Keynote Invited Session Chair: Walt DeGrange, CANA Advisors, Chapel Hill, NC, wdegrange@canallc.com 1 - Can Prediction be Better than Cure? On Analytics In Health-Care Edmund Jackson, Clinical Services Group, HCA, Nashville, TN, United States, edmund.jackson@hcahealthcare.com Healthcare is different: the intrinsic complexity, absolute moral imperatives and regulatory oversight of this business are unique. As such many of the technologies in healthcare differ from other industries. That said, the industry is entering a new regime where data is widely available, technology exists for analytics to run in real-time and the intention of bringing this intelligence into the workflow is widespread. Moreover, the advent of techniques such as diagnostic, predictive, and prescriptive analytics in other industries have ready applications in healthcare. The potential benefits of these activities to all stakeholders in the healthcare system, such as patients, providers and payers are enormous. In this talk Dr Edmund Jackson, Vice President and Chief Data Scientist of HCA will discuss this topic and provide a perspective of what has already been achieved and what is soon to come. Keynote Wednesday

395