INFORMS 2021 Program Book

INFORMS Anaheim 2021

WD20

CC Room 201D In Person: Interpretable Machine Learning General Session Chair: Cynthia Rudin, Duke University, Durham, NC, 27708, United States 1 - Multi-label Classification for Aviation Accident Reports Xinyu Zhao, Arizona State University, Tempe, AZ, 85281, United States, Hao Yan The NTSB aviation accident database recorded information about civil aviation accidents since 1982. The data are organized in a relational database where several tables jointly describe a specific accident. Finally, we end up with 61671 accidents in 62570 aircraft. Each accident is represented with a sequence of failure events which is very useful information for risk analysis. However, there exists inconsistency within the definition of the event among the dataset which prevents researchers from making full use of the dataset. In this project, we convert the problem into a multi-label classification problem which enables us to extract the consistent events automatically. Furthermore, the developed algorithms are able to identify the important sentences through the attention mechanism. Dimension reduction (DR) techniques such as t-SNE, UMAP, and TriMap have demonstrated impressive visualization performance on many real world datasets. One tension that has always faced these methods is the trade-off between preservation of global structure and preservation of local structure: these methods can either handle one or the other, but not both. In this work, our main goal is to understand what aspects of DR methods are important for preserving both local and global structure: it is difficult to design a better method without a true understanding of the choices we make in our algorithms and their empirical impact on the lower-dimensional embeddings they produce. 3 - Identifying Influential Factors on Recipients’ Quality of Life After Lung Transplantation using Predictive Analytics and Explainable AI Mostafa Amini, Oklahoma State University, Stillwater, OK, 74075, United States, Ali Bagheri, Dursun Delen Algorithmic modeling’s prediction power is crucial in healthcare systems where the patients’ lives are at stake. We employ predictive analytics and Explainable Artificial Intelligence (XAI) techniques to address the end-stage lung failure which leaves the patients with no option other than a transplantation. We rely on the UNOS (united network for organ sharing) data with a massive number of features associated with the donors, patients, and conditions in which the transplant is performed. We investigate the most influential factors on the prediction of the quality of life of the lung recipients which in turn participate in the utility function corresponding to the assignment of organ-patient. WD19 CC Room 203A In Person: Modern Approaches to Pricing General Session Chair: Michael L Hamilton, University of Pittsburgh, Morristown, NJ, 07960-5148, United States 1 - Loss Functions for Data-driven Personalized Pricing Max Biggs, Assistant Professor, University of Virginia, Charlottesville, VA, 02139-1784, United States, Ruijiang Gao, Wei Sun We study a pricing setting where each customer is offered a personalized price based on customer and/or product features that are predictive of the customer’s valuation for that product. Often only historical sales records are available, where we only observe whether each customer purchased a product at the price prescribed rather than the customer’s true valuation. As such, the data is influenced by the historical sales policy which introduces difficulties in estimating revenue from pricing policies. We approach this problem using ideas from causal inference and machine learning. In particular, we study how to formulate loss functions which directly optimize revenue, rather than going through an intermediate demand estimation stage. These loss functions have certain asymmetries which aren’t present in typical classification loss functions. 2 - PaCMAP: A New Algorithm for Dimension Reduction Cynthia Rudin, Duke University, Durham, NC, 27708, United States

2 - Revenue Management with Product Retirement and Customer Selection Harsh Tarak Sheth, Columbia University, New York, NY, 10027- 4052, United States, Adam Elmachtoub, Vineet Goyal We consider a multi-product revenue management problem where a seller has a fixed inventory of each product to sell to a set of customers. The seller sequentially offers the set of available products to the customers and can also choose to retire products at any point. Once a product is retired, it is no longer offered to any subsequent customers. When customers follow a common MNL choice model, we provide an asymptotically optimal policy for product retirement. When there are multiple customer types, we provide a policy for jointly selecting customers and retiring products that guarantees one fourth of the optimal policy. With multiple customer types and two products, we provide an asymptotically optimal policy. 3 - Feature-based Market Segmentation and Pricing Michael L. Hamilton, University of Pittsburgh, Pittsburgh, PA, 07960-5148, United States With the rapid development of data-driven analytics, many firms have begun experimenting with personalized pricing strategies, i.e. strategies that predict a customer’s valuation then offer them a individualized price. Ideally, a firm would perfectly predict each customer’s valuation and price their goods accordingly. Unfortunately, in practice these valuations are often predicted by the firm using noisy regression models, and the number of prices the firm can offer are constrained by operational considerations. In this work, we propose and analyze a general framework for semi-personalized pricing strategies where the seller uses features about their customers to segment their market, and where customers are offered segment level prices. WD20 CC Room 203B In Person: Health Care, Modeling and Optimization I Contributed Session Chair: Alireza Farnoush, Auburn University, Auburn, AL, 36830-3141, United States 1 - Analysis of Covid-19 Spread in the Metropolitan Areas Within U.S via Integrating Multi-source Data Bilal Majeed, University of Houston, Houston, TX, United States, Jiming Peng, Ying Lin, Li Ang The COVID-19 has wreaked havoc upon the world with over 162 million confirmed cases and a death toll of over 3.36 million. It is alarming that the United States contributes to about a quarter of these confirmed cases and deaths. In this talk, we analyze major metropolitan areas (MSAs) in the U.S. and compare MSAs with similar demographic characteristics, to explore the association between some COVID-19 related measurements and the demographic characteristics in MSAs. Particularly, we explore possible reasons for the MSAs with high mortality rate (MR) and fatality rate (FR). 2 - Exploring a Multi-tiered Priced Pediatric Vaccine Gavi Market via OR Through a hypothetically coordinated vaccine market framework, we explore how to enhance affordability in pediatric vaccine purchases in the Gavi market via challenging its single-price policy. First, we develop an innovative approach to estimate the unknown vaccine reservation prices at each of the Gavi market countries from their known single-tier market prices. Then, we use these estimates to determine the reservation prices for new market configurations in multi-tiered priced Gavi markets. Via an optimization-based process, we maximize the affordability of the different configurations. 3 - Optimal Deferral Strategies for Elective Surgeries with Application to Non-infected US Population. Kristian Singh, Benefits Science Technology, Needham, MA, United States, Omid Nohadani The time span between diagnosis and the actual surgery is often driven by non- medical reasons, as magnified by capacity constraints caused by COVID-19 pandemic. Determining this wait time is a decision in which physicians, patients, employers, and insurers all participate. Using claims data from a large portion of the US population who received a hernia surgery, we developed a framework consisting of a machine learning method to assess individual impact of delays and an optimization model to inform deferral strategies for surgeries. This framework, delivered through a web-based application, provides decision tools for tradeoffs of the consequences when deferring any given individual. Ruben Proano, Associate Professor, Rochester Institute of Technology, Rochester, NY, United States, Galo Mosquera

169