2016 INFORMS Annual Meeting Program

POSTER SESSION

INFORMS Nashville – 2016

Multi-agent Routing In Shared Guide Path Networks Greyson Daugherty, Georgia Institute of Technology, 765 Ferst Dr NW, Atlanta, GA, 30318, United States, sdaugherty3@gatech.edu, Spyros Reveliotis, Greg Mohler This poster describes a heuristic algorithm for minimizing the makespan required to route a set of agents inhabiting a shared guidepath network from their initial locations to their respective destinations. The work is motivated by operations taking place in the context of some unit-load material handling systems like zone- controlled AGV systems, as well as in quantum computers. This poster presents a brief description of the considered problem and of the inner workings of the proposed algorithm, along with a set of computational results that reveal the efficacy of the derived solutions and directions for further research. Cost Minimization Of Government Issued Cell Phones John Yannotty, Slippery Rock University, 1 Morrow Way, Slippery Rock, PA, 16057, United States, jcy1001@sru.edu, Alexander Reid Barclay Increasing costs associated with DSL circuits has led the United States Pentagon to study consolidation of its wireless network in attempt to minimize annual expense while maximizing efficiency. During consolidation, Private Virtual Channels (PVCs) are transferred from either low utilized or expensive circuits to existing circuits with higher utilization and lower annual cost. The consolidation process is further constrained by region, service package (VCI code), and utilization capacity per circuit. Through the use of an optimized network annual expenses are decreased from approximately $11 million to $3.1 million. Visualization Of Cross Validation For Prediction And Classification Alexander Engau, Associate Professor, University of Colorado Denver, P.O. Box 173364, Department of Mathematical Sciences, Denver, CO, 80217-3364, United States, aengau@alumni.clemson.edu, Paola Andrea Gonzalez In supervised learning, the performance of data mining and machine learning algorithms is often measured and compared only numerically using cross validation. Here, we describe two case studies in which we can complement and further extend such numerical comparisons with a visualization in the form of new validation maps. These validation maps are illustrated using several state-of- the-art classifiers from Scikit-Learn and offer substantial new insights into advantages and remaining limitations of support vector machines, decision trees, boosting and discriminant analysis. Bayesian Optimization Of Predictive Precision For Business Ledger Analytics Abhinav Maurya, Carnegie Mellon University, 5634 Stanton Avenue, Apt 306, Pittsburgh, PA, 15206, United States, ahmaurya@gmail.com, Aly Megahed Predicting changes in account revenues is of vital importance to a business in order to take action on accounts that are predicted to shrink, and to learn from success stories of offerings that led to maximum revenue growth. However, the corresponding datasets are often imbalanced, and therefore accuracy is a poor metric to optimize for. We present a Gaussian Process-based method that maximizes precision, yielding actionable results without sacrificing much accuracy. We find that our method gives better results than exhaustive uniform grid search, since Gaussian Process-based optimization can focus on areas of parameter space that have higher chances of attaining the maximum objective value. The Continuous Network Location Problem For The Alternative Fuel Refueling Station Sang Jin Kweon, PhD Candidate, Pennsylvania State University, Unstable price of oil and concerns about the finite nature of reserves, coupled with increasing awareness of the environmental issues caused by the burning of fossil fuels, increased spotlight on alternative-fuel vehicles. In order to stimulate the use of alternative-fuel vehicles, the inherent problem with the lack of refueling infrastructure must be resolved. In this study, we propose a novel methodology to locate an alternative-fuel refueling station on a road network with the objective of maximizing the total traffic flow covered by the station, so that more customers are able to refuel their alternative-fuel vehicles. Optimal Number Of Choices In Rating Contexts Sam Ganzfried, Assistant Professor, Florida International University, 11200 SW 8th St, Miami, FL, 33199, United States, sam.ganzfried@gmail.com In many settings people give numerical scores to entities from a small discrete set, e.g., attractiveness from 1-5 on dating sites and papers from 1-10 for conferences. We study the problem of understanding when using a different number of options is optimal. We study several natural processes for score generation. One may expect that using more options always improves performance, but we show that this is not the case, and that using fewer choices—even just two—can surprisingly be optimal. Our results suggest that using fewer options than typical could be optimal in certain situations. This would have many potential applications, as settings requiring entities to be ranked by humans are ubiquitous. 310 Leonhard Building, Industrial and Manufacturing Engineering, University Park, PA, 16802, United States, svk5333@psu.edu

Improving Patient Access To Primary Care Through E-visits Xiang Zhong, University of Florida, Gainseville, FL, 32601, United States, oliver040525@gmail.com, Jingshan Li, Philip Bain, Albert Musa, Peter Hoonakker To improve primary care access, many healthcare organizations have introduced electronic visits to provide patient-physician communication through securing messages. In this study, we introduce an analytical model to characterize primary care physician’s operations on office and e-visits, and other non-direct care tasks. Analytical formulas to evaluate the mean and variance of patient office visit and e-visit cycle times are derived, and discussions of the impacts of e-visits on traditional primary care delivery are carried out. It can be observed that the patient e-visit to office visit referral ratio plays an important role in determining whether it’s beneficial to conduct e-visits. Optimizing Inventory Under Non-stationary Demand For Profitability Improvement Liu Yang, Assistant Professor, Purdue University, 3000 Technology Ave, New Albany, IN, 47150, United States, LYang@purdue.edu This research presents a multi-period optimization model that integrates inventory classification and control decisions to maximize the NPV of profit. The model explicitly addresses nonstationary demand, limited inventory budget, arbitrary reviews periods, and SKU-specific lead times and holding costs. The model is applied to a real-life company that currently uses the multi-criteria inventory classification, and improves its profit by nearly 3%. The comparison to the ABC shows an average profit increase of 7.5%. We find that profit is insensitive to the number of classes in a wide range, but when the budget is tight, a large number of classes with a wide range of service levels is optimal. Analytics In Action: When Will I Get Out Of The Hospital? Modeling Length Of Stay Using A Disease Network Pankush Kalgotra, PhD Candidate, Oklahoma State University, 308 W Maple Avenue, Apt 5, Stillwater, OK, 74074, United States, pankush@okstate.edu, Ramesh Sharda Comorbidity is a medical condition when a patient develops multiple diseases simultaneously. We examine the impact of comorbidity on the patient’s hospital length of stay (LOS). We present a unique approach to model comorbidities by creating a network of diseases from the pair-wise combinations of the diseases diagnosed in the 1.6 million patient visits in US hospitals in 2011. Using the small-world property of the network, we proposed a new comorbidity score for a patient. Finally, we built an explanatory and predictive model on the patient visits in 2012, to predict a patient’s LOS. The model with our proposed comorbidity score outperformed the existing models. The GetFruved Project Uses Integer Programming To Match Freshmen To Peer Mentors Wangcheng Yan, The University of Tennessee, Knoxville, Knoxville, TN, 37996, United States, wcyan2009@hotmail.com, Wenjun Zhou, Sarah Colby, Kendra Kattelmann, Anne Mathews, Melissa D. Olfert In this study, we took a quantitative approach to the friend-matching problem to assign peer mentors (PMs) to freshmen (FMs) recruited for the GetFruved project. A 20 “fun”-question survey was used to develop the PM-FM matching algorithm. Data were collected from two semesters. Semester 1 served as training period, and matching was made in Semester 2. Our strategy was to train a model with logistic regression, and optimize the matching with integer programming. The empirical study verified the homophily theory, and demonstrated the effectiveness of our approach to identifying the optimal PM assignments. Dynamic Model Validation Metric Based On Wavelet Thresholded Signals Andrew D Atkinson, Captain, Air Force Institute of Technology, Model validation is a vital step in simulation development to ensure that a model is sufficiently representative of the system. Transient phase model validation deserves special attention because the experimental system data is often contaminated with noise, due to the short duration and sharp variations in transient data. We propose a process to validate the transient phase of a model that uses wavelet thresholding to de-noise the data signals and calculates a validation metric that incorporates shape, phase, and magnitude error. A simulation study and empirical data from an automobile crash study illustrate the wavelet thresholding validation approach. Joint Service In Primary Care Clinics Hyo Kyung Lee, University of Wisconsin-Madison, 313 N Frances To improve patient flow and reduce provider workload, joint service has been proposed and implemented in many primary care clinics. As no model is available yet to quantify joint visit’s impact, we introduce Markov chain models of patient flow with joint visits, and investigate the system behavior under different scenarios. Particularly, to reduce the state space dimension, convergent iterative procedures are proposed. Furthermore, the study is extended to non-Markovian case by introducing empirical formulas. To illustrate the applicability of the methods, an application study at Dean East Clinic is presented. 2950 Hobson Way, Wright-Patterson AFB, OH, 45433, United States, andrew.atkinson@afit.edu, Raymond R Hill St Apt 601, Apt 601, Madison, WI, 53703, United States, hlee555@wisc.edu, Xiang Zhong, Jingshan Li, Albert Musa, Philip Bain

301