Informs Annual Meeting 2017

MA38

INFORMS Houston – 2017

MA39

2 - Understanding Smart Service Systems through Text Mining Chiehyeon Lim, UC Merced; POSTECH, 790-784, Engineering Building #4-316, San 31, Hyoja-Dong Nam-Gu, Gyungbuk, Korea, Republic of, chlim@unist.ac.kr, Paul Maglio Despite the widespread application and research on smart service systems, understanding of such systems is limited. We analyze text data related to smart service systems to develop a unified understanding. Using a combination of metrics and machine learning algorithms, we identify important features, 16 research topics, 4 technology factors (connection, collection, computation, and communications), and 13 application areas from 5,378 scientific articles and 1,234 news articles. We also develop conceptual frameworks of smart service system based on the analytics result. Our contribution is to establish common ground for understanding smart service systems based on a data-driven approach. 3 - A Ranking Algorithm for Shipping in the Spot Market Max R. Biggs, Massachusetts Institute of Technology, Cambridge, MA, 02139, United States, maxbiggs@mit.edu, Georgia R. Perakis We study a problem in the tramp shipping industry where a broker has to choose from a set of available cargoes but does not know what cargoes will be available for subsequent voyages. Solving the naïve dynamic programming formulation is hard for realistic problem sizes due to the number of available cargo combinations being exponential in nature. Instead we propose a ranking algorithm that is polynomial in the number of cargoes available, and prove this is equivalent to solving the DP optimally. We run simulations to show that the ranking algorithm outperforms the other common heuristics by a minimum of 5% and the current market performance by 4-32%. 352C Complex Multi-criteria Decision Making Problems 1 Sponsored: Multiple Criteria Decision Making Sponsored Session Chair: Murat Mustafa Koksalan, Middle East Technical University, Middle East Technical University, Ankara, 06531, Turkey, koksalan@metu.edu.tr Chair: Banu Lokman, Middle East Technical University, Middle East Technical University, Ankara, 06800, Turkey, lbanu@metu.edu.tr 1 - Distribution-based Representative Sets for Multi-objective Integer Programs Banu Lokman, Middle East Technical University, Department of Industrial Engineering, Universiteler Mahallesi, Ankara, 06800, Turkey, lbanu@metu.edu.tr, Sami Serkan Ozarik, Murat Mustafa Koksalan Finding a set of points that represent the nondominated set for Multi-objective Integer Programs (MOIPs) is an important research area and has many applications. In this study, we search for common properties of the distributions of nondominated points in various MOIPs and show that the distribution of nondominated points may be critical in defining the desired properties of the representative subset. We introduce a density measure and analyze typical distributions of nondominated points for different MOIPs. We categorize the nondominated set into regions based on their estimated densities. We then MA38 Gokhan Ceyhan, Middle East Technical University, Orta Dogu Teknik Universitesi, Endustri Muhendisligi Bolumu No 319, Ankara, 06800, Turkey, gokhanceyhan01@gmail.com, Murat Mustata Koksalan, Banu Lokman, Murat Mustata Koksalan In this study, we will first briefly present our approaches to generate all or representative nondominated points for multi-objective mixed integer linear programming (MOMILP) problems. Then, we will demonstrate our web application where users can input their MOMILP problems, select different solver options based on the type of the output required and use different visualization tools available. 3 - GoNDEF: An Exact Algorithm to Find Non-dominated Solutions of Multi-objective MILP as Points, Line Segments and Facets Seyyed Amir Babak Rasmi, PhD Student, KOC University, Rumeli feneri st, KOC University, ENG212, Istanbul, 34450, Turkey, srasmi14@ku.edu.tr, Metin Turkay The Generator of Non-Dominated and Efficient Frontier (GoNDEF) finds exact Pareto solutions represented as points, line segments, and facets consisting of Extreme/non-Extreme Supported/Unsupported Non-dominated solutions for Multi-Objective MILPs. A large group of real-life decision problems involving more than one conflict criterion can be presented as MOMILPs. GoNDEF applies the Screening and the Generative Approaches to extract the non-dominated frontier from a given set of integer solutions and without a given set of integer solutions, respectively. generate distribution-based representative sets for different MOIPs. 2 - A Web Application for Solving Multi-objective Integer Programming Problems

352D Emerging Concepts II

Sponsored: Analytics Sponsored Session 1 - Deep Reinforcement Learning Daewoo Chong, Booz Allen Hamilton, Washington, DC, Machine intelligence is becoming more and more prevalent in our everyday lives. Andrew Ng, a leader in the field, calls it the new electricity. Many of the gains have come from a family of methods including convolutional and recurrent neural networks called deep learning. These methods have become crucial to the improvements in digital assistants, machine translation systems, and self-driving cars, and this seems to be only the beginning. Join this talk on deep reinforcement learning to learn about how using deep convolutional neural networks on games may be able transform the automation industry. 2 - Is Analytics in Professional Sports Making a Difference? Walter DeGrange, CANA Advisors, Chapel Hill, NC, wdegrange@canallc.com This presentation covers the impact of analytics on team performance in the top four sports in the United States (Major League Baseball (MLB), National Football League (NFL), National Basketball Association (NBA) and the Nation Hockey League (NHL)) over the past three years. Several analytics projects with professional teams and the issues with implementing solutions are also presented. Projects include developing models for an NHL team and a Major League Lacrosse (MLL) team. 352F Health Care, Modeling and Optimization Contributed Session Chair: Buyannemekh Munkhbat, University of Massachusetts-Amherst, Amherst, MA, United States, bmunkhbat@umass.edu 1 - Characterizing Obstructive Sleep Apnea a Descriptive Analytics Approach Rupesh Agrawal, Research Assistant, Oklahoma State University, 700 N. greenwood ave., Rm#313, Tulsa, OK, 74106, United States, rupesh.agrawal@okstate.edu, Dursun Delen, Bruce Benjamin Increasing adoption of EHR/EMR, propagated via affordable healthcare act of 2009, have opened new opportunities to improve health and healthcare outcomes. Obstructive sleep apnea (OSA) is a very common health condition, the existence of which would increase manifestations of deadly diseases such as heart failure. Despite proven adverse effects, the true understanding of the OSA is still in nascent stages. Hence, using a large and feature-rich EHR dataset, this analytics study aimed at characterizing OSA by identifying and analyzing associated comorbidities, and also developing a comprehensive taxonomy of OSA and related health conditions. 2 - The Impact of Health Information Exchange on Patient Mobility and Hospital Competition Xinxin Guo, Southeast University, 2 Sipailou, Nanjing, 210096, China, bzhmgx@163.com, Haiyan Wang The potential benefit of health information exchange (HIE) for patients is that reduces the costs patients face to switch providers. This paper examines the impact of the adoption of HIE on health care quality and hospital profits as patients, especially those experienced patients, have the freedom to choose the provider they receive treatment. We develop a two-period duopoly model in which health care quality is differentiated and patients are heterogeneous. We give conditions under which the adoption strategy benefits or hurts providers and social welfare. 3 - Building Multiscale Models for Predicting Spatial Changes between Serial Functional Images of Lung Cancer Patients Chunyan Duan, Tongji University, Room1114, Tongji Building Block A, No. 1 Zhangwu Road, Shanghai, 200092, China, duanchunyan77@163.com Chunyan Duan, University of Washington, 1959 NE Pacific St. MA41

Box 356043, Seattle, WA, 98105, United States, duanchunyan77@163.com, Stephen R. Bowen, W. Art Chaovalitwongse, Daniel S. Hippe, Daehan Won, Jianxin You, Jianxin You

Lung cancer response varies spatially during therapy. We are modeling serial functional cancer imaging changes from statistically independent regions within lung tumors by cluster analysis and spatial autocorrelation. Features of multiscale cancer regions are trained, cross-validated, and tested in our models to predict spatial changes during and after therapy. This represents a powerful approach to assess local therapeutic efficacy.

146