2015 Informs Annual Meeting

SA37

INFORMS Philadelphia – 2015

SA37 37-Room 414, Marriott Health Care Modeling and Optimization I Contributed Session Chair: Michal Jakubczyk, Warsaw School of Economics, Al. Niepodleglosci 162, Warsaw, 02-554, Poland, michal.jakubczyk@sgh.waw.pl 1 - The Optimal Timing of Medical Tests Thomas Weber, Associate Professor, EPFL, CDM-ODY 3.01, Station 5, Lausanne, VD, 1015, Switzerland, thomas.weber@epfl.ch This paper considers the optimal timing of tests based on a known law of motion for the statistical evolution of a random population prevalence. In a Bayesian setting, we find for a given imperfect binary disease diagnostic and action thresholds the optimal time to test and retest a potentially ill individual, conditional on past test outcomes. The framework allows for complex disease dynamics, including multiple populations, contagion, and stochastic lifetimes. 2 - A Fuzzy Approach to Modeling the Willingness-to-Pay for Health and Supporting Decision Making Michal Jakubczyk, Warsaw School of Economics, Al. Niepodleglosci 162, Warsaw, 02-554, Poland, michal.jakubczyk@sgh.waw.pl Choosing between health technologies involves multiple criteria (e.g., effects & cost), uncertainty, and often multiple alternatives. I advocate that, due to peculiarity of health, fuzziness additionally needs to be introduced to model the willingness-to-pay/accept (WTP/WTA). I show how to do that by redefining notions typically used in health technology assessment. Properties of new approach are discussed. Accounting for fuzziness additionally explains the WTP- WTA disparity in this context. 3 - Emergency Department Length-of-stay Estimation using Time- variant Predictors Seung Yup Lee, Graduate Research Assistant, Wayne State University, 4815 4th St., Detroit, MI, 48202, United States of America, seung.lee@wayne.edu, Ratna Babu Chinnam, Alper Murat, Evrim Dalkiran The accurate length-of-stay (LOS) estimation for patients in emergency departments (ED) is a pre-requisite for quality resource coordination between ED and inpatient wards. We investigate how time-variant levels of crowding in ED can be captured and incorporated in LOS estimation models by using vector autoregression (VAR). We will also report results and insights from testing the models on data from VA Medical Centers. 4 - An Integrated Framework to Model the Trajectories of Chronic Conditions Adel Alaeddini, University of Texas at San Antonio (UTSA), One UTSA Circle, San Antonio, United States of America, adel.alaeddini@utsa.edu Any medical condition that requires long term monitoring and management to control symptoms and shape the course of the disease is known as chronic conditions. Nearly 45% of the general population has 1 chronic condition or more. This accounts for more than 75% percent of health care expenditures. We present an integrated probabilistic framework for modeling the trajectories of chronic conditions. The proposed methodology will be validated using a large dataset from a medical center in Texas. SA38 38-Room 415, Marriott Big Data I Contributed Session Chair: Ellick Chan, Exponent, 149 Commonwealth Dr., Menlo Park, CA, 94025, United States of America, echan@exponent.com 1 - A Structural Service Model for Describing and Designing Services with Data Chie-Hyeon Lim, Post-doc, POSTECH, Engineering Building #4-316, Pohang, 790-784, Korea, Republic of, arachon@postech.ac.kr, Min-Jun Kim, Kwang-jae Kim, Paul Maglio Using big data effectively in service design requires having a model that describes the service in question along with the data in use. In this talk, we propose a generic structural service model to describe a service with a set of predefined variables, facilitating design of services that use big data. The variables include service objective, indicators, customer and context variables, and delivery contents. We discuss the model in the context of several case studies of service design.

2 - Increasing Productivity and Minimizing Errors in Spreadsheet Analytics Larry LeBlanc, Professor, Owen Graduate School of Management,

Vanderbilt University, 401 21st Avenue South, Nashville, TN, 37203, United States of America, larry.leblanc@owen.vanderbilt.edu, Thomas Grossman, Michael Bartolacci

Spreadsheets have proliferated for business analytics, and spreadsheet errors can result in poor supply chain, manufacturing, or investment decisions, including the failure to identify good opportunities. We examine potential problem areas for spreadsheet design and suggest alternative design approaches that seek to increase productivity and reduce the likelihood of errors. Even careful analysts might send their spreadsheet to assistants for updating, and s/he might need these guidelines 3 - A Practical Big Data Precision Marketing – Cross-Selling Mobile Bank to Internet Bank Jian Xu, IBM, Diamond Bld, ZGC Software Park, Beijing, China, xujianx@cn.ibm.com, Ming Xie, Yuhang Liu, Zhen Huang, Tianzhi Zhao, Yuhui Fu The bank wants improve mobile bank users and transform customers from online bank channel to mobile bank. Mobile bank represents the future E-channel. Large amount of data is integrated and analyzed on E-channel users’ behavior. The users’ online behaviors are also considered. We build the cross-selling model to identify the potential customers who are more likely to become mobile bank users, and improve the marketing success rate significantly. 4 - Forecasting Unemployment Rate by using Ensemble Hybrid Ann- Bayesian Model Combination Farzad Radmehr, West Virginia University, 900 Willowdale Road, Morgantown, WV, United States of America, fradmehr@mix.wvu.edu The goal of this paper is to predict the future data by using ensemble Bayesian model. Our dataset is UK unemployment rate from Floros C. paper in 2005(Floros, 2005). In this paper, the Bayesian Ensemble Model Combination (BMC) will be proposed. For this purpose, we run ANN multiple times and these results will be the initial values for BMC. Then by giving the weight to each value, we predict the new value. The goal is to compare the values in BMC and ANN. 5 - Deep Learning Approaches to Digging Data Out of Digitized Paper Documents Ellick Chan, Exponent, 149 Commonwealth Dr., Menlo Park, CA, 94025, United States of America, echan@exponent.com, Glen Depalma Many organizations scan paper documents for fast search, however, existing search approaches generally require carefully crafted search terms to find documents. In this talk, we discuss deep learning approaches for OCR and search. We use computer vision to improve OCR accuracy and apply deep learning using Google’s Word2Vec natural language processing (NLP) to identify topics of interest automatically. We’ve processed more than 300 boxes of documents with our techniques. SA39 39-Room 100, CC Game Theoretic Models in Operations and Marketing Interface Cluster: Operations/Marketing Interface Invited Session Chair: Tao Li, Santa Clara University, 500 El Camino Real, Santa Clara, CA, 95053, United States of America, tli1@scu.edu 1 - Online Manufacturer Referral to Heterogeneous Retailers Gangshu Cai, Santa Clara University, OMIS Department, Lucas Hall 216N, Santa Clara, CA, 95053, United States of America, gcai@scu.edu, Hao Wu, Chwen Sheu, Jian Chen Since the development of the Internet, thousands of manufacturers have been referring consumers visiting their websites to some or all of their retailers. Through a model with one manufacturer and two heterogeneous retailers, we investigate whether it is an equilibrium for the manufacturer to refer consumers exclusively to a retailer or nonexclusively to both retailers. 2 - Strategic Risk Management in Spot Market for Supply Chains under Competition Xuan Zhao, Associate Professor, Wilfrid Laurier University, 75 University Avenue West, Waterloo, ON, Waterloo, Canada, xzhao@wlu.ca, Shanshan Ma, Wei Xing This paper studies two risk management strategies related to spot market to mitigate firms’ exposure to demand uncertainty, namely, operational hedging and financial hedging. We provide insights on the dynamics of each hedging strategy under competition.

50