2015 Informs Annual Meeting

SB38

INFORMS Philadelphia – 2015

SB39 39-Room 100, CC Interdisciplinary Focus on Problem Solving Cluster: Operations/Marketing Interface Invited Session

3 - Early Warning Methods and Predictive Models for Hospital Risk and Readmissions Jakka Sairamesh, Ceo And President, CapsicoHealth, Inc, 2225 E Bayshore Rd. Ste. 200, Palo Alto, CA, 94303, United States of America, ramesh@capsicohealth.com, Ruichen Rong This poster and research abstracts presents the effectiveness of methods for improving patient quality outcomes (e.g. reducing 30-day readmissions) based on clinical and cost based factors. We will present early-warning methods to predict patients at risk of 30-day readmissions based on past admissions, ER visit rates, mortality rates, and charges. The dominant factors includee clinical risk, costs, emergency room visits and mortality rates. The prediction showed nearly 88 percent accuracy. SB38 38-Room 415, Marriott Big Data II Contributed Session Chair: Christoph Wunck, Professor, Jade University of Applied Sciences, Friedrich-Paffrath-Str. 101, Wilhelmshaven, 26389, Germany, wunck@jade-hs.de 1 - Making Billions of Decisions a Day: Experiment Driven Bid Optimization in Online Advertising Daizhuo Chen, PhD Candidate / Senior Data Scientist, Columbia Business School / Dstillery, 470 Park Avenue S, 6th Floor South, New York, NY, 10016, United States of America, dchen16@gsb.columbia.edu, Robert Phillips, Garrett Van Ryzin, Brian D’alessandro, Perlich Claudia Online advertising is a good playground for operations researchers who want to explore the opportunities and pitfalls of the “Big Data” promise. This talk will focus on bid optimization: as an ad buyer, how to determine the best bid prices for billions of ad opportunities everyday, in today’s dynamic and opaque marketplaces of real-time bidding. We introduce a solution based on continuous experimentation and optimization, and touch on a paradox of big data: you never have the data you need. 2 - Fast Gaussian Process Regression for Large Non-Stationary Spatial Data Babak Farmanesh, Oklahoma State University, 322 Engineering North, Stillwater, OK, 74074, United States of America, babak.farmanesh@okstate.edu, Arash Pourhabib We propose Sparse Pseudo-input Local Kriging (SPLK) as a predictive model for large non-stationary spatial data. SPLK uses orthogonal cuts to create small domains, where it applies sparse Gaussian process regression. The orthogonal cuts enable SPLK to be applied to spatial datasets that include exogenous variables, hence having a dimension greater than three. We apply SPLK to real and simulated datasets and demonstrate it can efficiently predict the response variable. 3 - Advanced Decision-Making Procedures in Massive Failure Data Classification Keivan Sadeghzadeh, Graduate Research Assistant, Northeastern University, 27 Payne Rd, Newton, MA, 02461, United States of America, k.sadeghzadeh@neu.edu, Nasser Fard In many professional areas, management decision-making process is based on the type and size of data where data classification is a necessary procedure. Massive amount of data in high-dimensions are increasingly accessible from various sources and it has become more difficult to process the streaming data in traditional application approaches. This paper presents advanced procedures to analyze high-dimensional failure data in order to facilitate decision-making through data classification. 4 - Preprocessing of Manufacturing Process Signals using Wavelet-based Filters Christoph Wunck, Professor, Jade University of Applied Sciences, Friedrich-Paffrath-Str. 101, Wilhelmshaven, 26389, Germany, wunck@jade-hs.de Bringing Big Data to the shop floor is one of the current visions to increase product quality, process reliability and overall productivity in manufacturing. Extracting the information content hidden in the flood of sampled data of process variables in real time requires data reduction and filtering techniques that are not commonly applied in manufacturing processes. A case study on injection molding shows how wavelet-based data preprocessing can simplify any subsequent data analysis.

Chair: Kathleen Iacocca, University of Scranton, 439 Brennan Hall, Scranton, PA, United States of America, kathleen.iacocca@scranton.edu 1 - A Functional Robust Newsvendor Model with Uncertain Price- Sensitive Demand Junxuan Li, Graduate Student Instructor, University of Michigan- Dearborn, 23935 W Outer Dr. Apt. H16, Melvindale, MI, 48122, United States of America, junxuanl@umich.edu, Jian Hu, Sanjay Mehrotra A functional robust newsvendor model with coordination of pricing and inventory decisions in uncertain market is proposed, which specifies an uncertainty set of nonparametric demand curves and seeks the best decisions against the worst case. We discuss the impact of functional robustness and wholesale price and develop a cut generation algorithm for convex demand curve case, while reformulate the model as second-order conic program for concave case. A grocery store case study is then discussed. 2 - Supply Chain Robust Optimization: An Examination of the Relationship Between Flexibility, Agility Seyed Nooraie, NCAT university, 2205 New Garden Rd #1210,

Greensboro, NC, 27410, United States of America, snooraie@aggies.ncat.edu, Mahour Mellat Parast, Paul M Stanfield

A robust multi-objective mixed integer nonlinear programming model is defined for an Aggregate Production Planning (APP) incorporating three conflicting objective functions simultaneously. We propose a theoretical construct linking elements of uncertainty with aspects of agility and flexibility where we use responsiveness to enhance these elements to overcome the future risk. 3 - Spreadsheet Approach for Integrating Production, Marketing, and

Finance Decisions in Aggregate Plans Kathleen Iacocca, University of Scranton,

439 Brennan Hall, Scranton, PA, United States of America, kathleen.iacocca@scranton.edu, Kingsley Gnanendran

The well-known aggregate planning model is extended to explicitly incorporate demand forecasts as modified by any sales promotions, along with working capital constraints. The components of the extended model are implemented on a spreadsheet and linked using VBA programming to demonstrate the ease of managerial applicability. Optimal decision variables include workforce size; production, inventory, and outsourcing levels; and the magnitude and timing of price promotions and infusions of cash. SB40 40- Room 101, CC Organization Theory II Contributed Session Chair: Amit Das, Associate Professor, Qatar University, P.O. Box 2713, Doha, Qatar, amit.das@qu.edu.qa 1 - Refocusing Business Incubator Research: The Process of Business Incubation Literature on business incubation resembles the early stages of entrepreneurship research. Research has focused on defining business incubators and the associated outcomes. The lack of a cohesive framework has resulted in numerous definitions that have prevented adequate theoretical development. Presented is a conceptualization of the incubation process and how various theories from the organizational sciences can inform future research. 2 - Performance in Knowledge Intensive Environments: Interplay of Worker and Managerial Experience Juan Pablo Madiedo, PhD Candidate, IE Business School, Calle Maria de Molina 12, Bajo, PhD Program Office, Madrid, 28028, David Scheaf, UNC Charlotte, 9201 University City Blvd, Charlotte, NC, United States of America, dscheaf@uncc.edu

Spain, jpmadiedo.phd2014@student.ie.edu, Aravind Chandrasekaran, Fabrizio Salvador

This study examines the importance of workers and managers as sources of experience in a knowledge intensive work environment. We analyze the interplay of workers and managers and its effect on task performance. We use a dataset with information on over 1500 software maintenance tasks collected from a global IT and consulting corporation for testing our model.

78