2015 Informs Annual Meeting

TD21

INFORMS Philadelphia – 2015

TD22 22-Franklin 12, Marriott Contact Centers Sponsor: Applied Probability Sponsored Session Chair: Rouba Ibrahim, University College London, London, N7 8EP, United Kingdom, rouba.ibrahim@ucl.ac.uk 1 - Telephone Call Centers: Asymptotic Optimality of Myopic Forecasting-scheduling Scheme Han Ye, University of Illinois at Urbana Champaign, 350 Wohlers Hall, 1206 South Sixth Street, Champaign, IL, 61820, United States of America, hanye@illinois.edu, Noah Gans, Haipeng Shen, Yong-Pin Zhou We determine workforce schedules for call center arrivals that are doubly stochastic. Period-by-period arrival rates follow a hidden AR(1) process, and only arrival counts are observed. We formulate stochastic programs to minimize long- run average staffing costs, subject to a long-run average constraint on abandonment. We show that, in steady state, repeated, myopic solution of the single-period problem is stable, has low cost, and meets the abandonment constraint. 2 - A Structural Model for Agents’ Strategic Behavior in Call Centers Dongyuan Zhan, University of Southern California, Los Angeles, CA, United States of America, Dongyuan.Zhan.2015@marshall.usc.edu, Amy Ward, Seyed Emadi We do an empirical study of agent behavior in call centers. We begin by observing that regression analyses have low explanatory power, even though the data shows that agents speed up or slow down depending on the system load and their fatigue level. This leads us to investigate utility based structural models for agent behavior. 3 - Capacity Sizing with a Random Number of Agents Rouba Ibrahim, University College London, London, N7 8EP, United Kingdom, rouba.ibrahim@ucl.ac.uk We study the problem of staffing many-server queues with general abandonment and a random number of servers. For example, uncertainty in the number of servers may arise in virtual call centers where agents are free to set their own schedules. We rely on a fluid model to determine optimal staffing levels, and demonstrate the asymptotic accuracy of the fluid prescription. We also characterize the optimal staffing policy with self-scheduling agents. TD23 23-Franklin 13, Marriott Markov Decision Models and Approximations for Manufacturing Cluster: Stochastic Models: Theory and Applications Invited Session Chair: Tugce Martagan, Eindhoven University of Technology, 5600 MB Eindhoven, Eindhoven, Netherlands, T.G.Martagan@tue.nl 1 - Robust Approximate Dynamic Programming and Structured Policies for Degradable Energy Storage Marek Petrik, IBM, 1101 Kitchawan Rd., Yorktown Heights, NY, 10598, United States of America, mpetrik@us.ibm.com Batteries hold great promise for energy storage in arbitrage in electric grids but can degrade rapidly with use. In this talk, we analyze the impact of storage degradation on the structure of optimal policies and describe robust approximate dynamic programming methods that take advantage of the policy structure. 2 - Component Reservation for Asymptotically Optimal Allocation in Assemble to Order Production Systems Haohua Wan, University of Illinois at Urbana-Champaign, 104 South Mathews Ave., Urbana, IL, United States of America, hwan3@illinois.edu, Qiong Wang Component reservation is not myopically optimal as it sometimes holds back components from existing demands. We prove that in many cases, without reservation, component allocation cannot be asymptotically optimal, i.e., the percentage difference of the discounted inventory cost from its lower bound does not converge to zero as demand and production volumes increase, even though such convergence is achievable under other policies that reserve components for high-value product demands.

5 - Credit Scoring using Dynamic State Space Model under Statistical Volatility Linna Du, Data Scientist, CACS, 2259 Adam Clayton Powell, New York, NY, 10027, United States of America, linna.du@gmail.com In emerging market where the credit score and credit history are not trustworthy, the estimation and prediction of the credit score and prepayment risks are very important. In the paper, we propose a dynamic state space model considering the volatility and dynamic feature of the lending market. We found that the time varying volatility model provides better prediction than other time series models. We also identify the key factors that drive the lending risks. Disease Modeling in OR Sponsor: Health Applications Sponsored Session Chair: Emine Yaylali, Senior Service Fellow, Centers for Disease Control and Prevention, 1600 Clifton Road, Atlanta, GA, 30333, United States of America, wqq3@cdc.gov 1 - The Potential Impact of Reducing Indoor Tanning on Melanoma Prevention in the United States Yuanhui Zhang, CDC, Chamblee GA 30341, United States of America,, yfp5@cdc.gov, Donatus Ekwueme, Sun Hee Rim, Meg Watson, Gery Guy More than 700,000 adults in the United States are treated for melanoma each year, resulting in annual direct medical costs of $3.3 billion dollars and 9,000 deaths. We developed a Markov model to estimate the health and economic impacts of reducing indoor tanning for melanoma prevention in the United States under certain assumptions. According to this model, reducing indoor tanning may result in favorable savings in medical costs and life-years, comparable to other national prevention efforts. 2 - Estimating the Impact of HIV Care Continuum Interventions on the Reproduction Number Yao-Hsuan Chen, CDC, Chamblee GA 30341, United States of America,xhj1@cdc.gov, Andrew Hill, Paul G. Farnham, Stephanie L. Sansom We used a compartmental model to study HIV transmission in the United States from 2006 through 2020 among heterosexuals, men who have sex with men, including bisexual men, and injection drug users. We analyzed the impact of interventions to improve HIV diagnosis, care, and treatment on the reproduction number. Analyses using this model can provide insights into the long-term effectiveness of HIV prevention strategies. 3 - Stratifying Risk Groups in Compartmental Epidemic Models: Where to Draw the Line? Margaret L. Brandeau, Professor, Stanford University, MS&E Department, Stanford, CA, 94305, United States of America, brandeau@stanford.edu, Jeremy D. Goldhaber-fiebert Disease models used to support cost-effectiveness analyses of health interventions are often stratified to reflect population heterogeneity (e.g., age, gender, risk behaviors). We examine the impact of population stratification in dynamic disease transmission models: specifically, the impact of different divisions of a population into a low-risk and a high-risk group. We show that the way in which the population is stratified can significantly affect cost-effectiveness estimates. 4 - Developing a Dynamic Compartmental Model of HIV in the United States Emine Yaylali, Senior Service Fellow, Centers for Disease Control TD21 21-Franklin 11, Marriott and Prevention, 1600 Clifton Road, Atlanta, GA, 30333, United States of America, wqq3@cdc.gov, Paul G. Farnham, Stephanie L. Sansom, Katherine A. Hicks, Emily L. Tucker, Amanda Honeycutt Over 1 million people in the US are living with HIV. To observe trends in HIV and evaluate the effectiveness of prevention interventions, we developed a dynamic compartmental model of disease progression and transmission. The population was stratified by age, sex, circumcision status, race/ethnicity, transmission group, and risk level. People progressed between compartments defined by disease status and care and treatment stage. Outcomes included HIV incidence, prevalence, and care status.

352