INFORMS 2021 Program Book

INFORMS Anaheim 2021

SD19

2 - Incorporating Learning-by-Doing Into Mixed Complementarity Equilibrium Models Benjamin D. Leibowicz, Assistant Professor, University of Texas- Austin, Austin, TX, 78712-1591, United States, Baturay Calci, Jonathan F. Bard, Gopika Jayadev Energy market equilibrium models are often specified and solved as mixed complementarity problems (MCPs). A limitation of existing MCPs is that they treat costs as exogenous input parameters. Therefore, MCPs have not been able to capture learning-by-doing (LBD), the empirically observed phenomenon whereby production costs tend to decline as a function of cumulative production experience. In this paper, we demonstrate the incorporation of LBD into a mixed complementarity equilibrium model. Through theoretical analysis and numerical exploration, we establish the conditions under which LBD formulations lead to convex optimization problems, which is important for inclusion in an MCP. Then, we demonstrate the practical application of a mixed complementarity equilibrium model with LBD using the North American natural gas market as an example. 3 - How Many Years of Data is Enough?: Using Multiple Years of Data to Increase Performance of Electricity System Models Tyler H. Ruggles, Research Scientist, Carnegie Science, Stanford, CA, United States, David J. Farnham, Nathan S. Lewis, Ken Caldeira Wind and solar generation are both subject to geophysical variability in their power output from hour to hour and on longer time scales. Recent studies looked at the frequency and duration of resource droughts that could have detrimental impacts of wind- and solar-powered energy systems. In this study, we use multi- decadal historical electricity demand and wind and solar data to study reliable least-cost electricity systems optimized over multiple years of input data. We show how asset capacities and system cost increase as a function of the number of years of input data (Nyrs). The performance of systems improves as a function of Nyrs when tested on out-of-sample years of data, which may help guide long- term system planning decisions. SD19 CC Room 203A In Person: Predictive Analytics Towards Improved Health Outcomes General Session Chair: Maryam Kheirandish, University Arkansas, AR, United States Co-Chair: Sasa Zorc, University of Virginia, Darden School of Business, Charlottesville, VA, 22903, United States 1 - Using Simple Optimization Methods to Enhance the Development of Stratified Models for Infectious Disease Anthony C. Nguyen, PhD Student, University of Southern California, Los Angeles, CA, 92886-9014, United States, Sze-chuan Suen Infectious disease models are powerful tools for assessing health policy benefits. It is critical to stratify these models by demographic characteristics, such as race/ethnicity and age, if prevalence, incidence, transmission, or treatment vary across these factors. However, this may be challenging as limited subpopulation- specific data may be available. We use simple optimization techniques to parameterize a race- and age-stratified model of HIV in Los Angeles given limited data from publicly available surveillance reports. Given these straightforward formulations, we hope these formulations can be widely adopted among modelers in public health and epidemiological disciplines. 2 - Feature Engineering for Opioid Overdose Prediction Shengfan Zhang, University of Arkansas, Fayetteville, AR, 72701, United States, Ryan Sanders, Maryam Alimohammadi We develop an efficient methodology for extracting features from time-dependent variables in transaction data. Transaction data is collected at varying time intervals making feature extraction more difficult. Unsupervised representational learning techniques are investigated, and the results compared with those from other feature engineering techniques. This methodology is then applied to insurance claims data in order to find features to predict whether a patient is at risk of opioid overdose. Features created are input to recurrent neural networks with long short-term memory cells. Hyperparameters are found through Bayesian optimization. Validation data features are reduced using weights from the best model and compared against those found using unsupervised learning techniques in other classifiers. 3 - Pricing the Covid-19 Vaccine: A Mathematical Approach Banafsheh Behzad, California State University-Long Beach, Long Beach, CA, 90808-0506, United States, Susan E. Martonosi, Kayla Spring Cummings We use optimization and game theoretic approaches to model the COVID-19 U.S. vaccine market as a duopoly with two manufacturers Pfizer-BioNTech and Moderna. The results suggest that even in the context of very high production and distribution costs, the government can negotiate prices with the manufacturers to keep public sector prices as low as possible while meeting

demand and ensuring each manufacturer earns a target profit. Furthermore, these prices are consistent with those currently predicted in the media.

SD20 CC Room 203B In Person: But, what if we were Wrong: Modeling and Analysis of Policy-related Strategies Against the COVID-19 Pandemic General Session Co-Chair: Dan Yamin, Tel Aviv University Iby and Aladar Fleischman Faculty of Engineering, Rehovot, Israel 1 - Early Detection of COVID-19 Outbreaks Using Human Mobility Data Grace Guan, Stanford University, Stanford, CA, United States, Yotam Dery, Matan Yechezkel, Irad Ben-Gal, Dan Yamin, Margaret Brandeau To appropriately target the timing, location, and severity of measures intended to reduce COVID-19 spread, it is essential to predict when and where outbreaks will occur, and how widespread they will be. We analyze aggregated, anonymized health and cell phone mobility data from Israel. We develop predictive models for daily new cases and the test positivity rate to predict the severity of COVID-19 in districts of Israel over the following week. Models using mobility data outperformed models that did not use mobility data. Our models accurately predict outbreak severity as defined by the tiers. Our method provides a useful tool for government decision makers. 2 - Human Mobility and Poverty as Key Drivers of COVID-19 Transmission and Control Matan Yechezkel, Tel Aviv University, Tel Aviv, Israel Applying heavy nationwide restrictions is a powerful method to curtail COVID-19 transmission but poses a significant humanitarian and economic crisis. We analyzed aggregated and anonymized mobility data from the cell phone devices of >3 million users in Israel. We integrated these mobility patterns into age-, risk- and region-structured transmission model. We found that localized and temporal interventions during regional outbreaks, that focus on individuals at risk, can substantially reduce mortality. Utilizing cellphone data despite being anonymized and aggregated can help policymakers worldwide identify hotspots and apply designated strategies against future COVID-19 outbreaks. 3 - A Multilayer Model for Early Detection of COVID-19 Dan Yamin, Tel Aviv University, Tel Aviv, Israel, Erez Shmeuli, Ronen Mansuri, Matan Porcilan, Tamar Amir, Lior Yosha, Matan Yechezkel, Tal Patalon, Sharon Handelman-Gotlib, Sivan Gazit We developed a machine-learning model, for COVID-19 detection that utilizes four layers of information: 1) sociodemographic characteristics of the tested individual, 2) spatiotemporal patterns of the disease observed near the testing episode, 3) medical condition and general health consumption of the tested individual over the past five years, and 4) information reported by the tested individual during the testing episode. Analyzing data of 140,682 individuals, our model obtained an area under the curve of 81.6%. Our ability to predict early the outcomes of COVID-19 tests is pivotal for breaking transmission chains, and can be utilized for a more efficient testing policy. SD21 CC Room 204A In Person: Empirical Studies in Healthcare Operations General Session Chair: Hyun Seok (Huck) Lee, Korea University Business School 1 - Machine Learning Methods to Predict Operational Surges in the Emergency Department During Covid-19 Ari J. Smith, University of Wisconsin Madison, Madison, WI, United States, Justin J. Boutilier, Manish Shah, Brian Patterson, Michael Pulia, Frank Liao The context of the Covid-19 pandemic has led to new perceptions of the factors that contribute to emergency department arrival and admission rates. We create machine learning models using recent department and patient-level data collected in an ED in conjunction with community Covid-19 metrics to predict the volume of arriving and admitted patients in short time frames. Predictions will motivate training activities and operational decisions in preparation of surges.

36