Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TD27

2 - Waterfall Bandits: Learning to Sell Ads Online Zheng Wen, Adobe Systems, 345 Park Avenue, E07-415, San Jose, CA, 95110-2704, United States, Branislav Kveton, Saied Mehdian, S. Muthukrishnan, Yikun Xian A popular model of pricing in online advertising uses what is known as the waterfall, where the publisher orders ad networks with their offered prices, and then contacts them in that order sequentially. A common approach to revenue maximization in the waterfall model is to estimate the model from past data and then maximize the revenue in the estimated model. This is statistically inefficient because the learning and optimization are separated. We propose an online learning algorithm for solving this problem, which interleaves learning and optimization. We also derive a sub-linear regret bound for the proposed algorithm, and evaluate it on both synthetic and real-world data. 3 - Data Scientist Professions in Big Data Era Marwah Halwani, King Abdulaziz University, Jeddah, 76208, Saudi Arabia, Nicholas Evangelopoulos, Victor Prybutok With the evolution of technology and big data, industry has adopted new job titles that require a new set of technical skills. This research examines the differences and commonalities among company-posted job requirements for five overlapping job titles, which include statistical analysts, big data analytics professionals, data scientists, data analysts, and business analytics professionals, by obtaining and analyzing online job descriptions through latent semantic analysis. The results are used to clarify skill requirements for big data professionals for the joint benefit of the job market where they will be employed, and academia that prepares such professionals. n TD27 North Bldg 132B Train the Trainer: Learn How to Bring O.R. to Your Local High School Sponsored: Education (INFORMED) Sponsored Session Chair: Theresa M. Roeder, San Francisco State University, San Francisco, CA, 94132, United States 1 - Train the Trainer: Learn How to Bring O.R. to Your Local High School Theresa M. Roeder, San Francisco State University, San Francisco State University, San Francisco, CA, 94132, United States Do you love O.R. and wish you had learned about it sooner? Do you wish you could introduce high school students to the fun math and stats applications our field can provide and answer the perennial question of “when am I ever going to use this”? Come to our Train the Trainer session, where we will provide you a toolkit with everything you need to organize a workshop for high school teachers; the workshop will introduce them to these real-world applications that they can use in their classrooms to engage students. n TD28 North Bldg 221A Maintenance Sponsored: Railway Applications Sponsored Session Chair: Nathaniel O. Richmond, BNSF Railway,The Colony, TX, 75056, United States Co-Chair: Steven Jay Tyber, General Electric, Chicago, IL, 60613, United States 1 - Predictive Switch Health Using Switch Amperage Casey Jen, CSX Corporation, Jacksonville, FL, United States Among all the Communication and Signal (C&S) components, switches cause biggest number of train delays on line of road. In this study, we leveraged switch amperage and developed a set of classification models to identify switch health. Taking information from multiple data sources, such as switch amperage, switch inspection records, switch incidents records, and many others, these classification models will enable CSX to proactively maintain our vital assets and better plan C&S workforce.

2 - Predicting Rail Defects with Massive Foot-by-foot Track Geometry Data Reza Mohammadi, University at Buffalo, Amherst, NY, United States, Faeze Ghofrani, Qing He Using foot-by-foot track geometry data, this study examines the relationship between rail defects and track geometry measurements. Fast Fourier Transformation is applied to extract the frequency-domain feature from track geometry data. To identify the significant variables in terms of variation, Singular Vector Decomposition (SVD) is implemented. In fact, SVD is applied to explain the source of variation in the dataset. Studying the behavior of rail defects based on this well-prepared track geometry measurements data set results in time-and cost-effective track predictive maintenance. 3 - Using Predictive Analytics for Railway Track Maintenance Planning Clark Cheng, Sr. Director Operations Res & Chief Data Scientist, Norfolk Southern Corporation, 1200 Peachtree Street NE, Atlanta, GA, 30309, United States, Mabby Amouie, Gongli Duan Railway track maintenance is critical to safety and efficient operations. It’s capital and labor-intensive. The planning process is manual and largely relies on the readings from the most recent single run of track geometry car. In this presentation, we will describe a machine-learning approach to predicting rail wear over five years based on historical track geometry car readings. The rail wear prediction model has significantly improved the track maintenance planning at Norfolk Southern. 4 - Integrated Optimization of Timetable and Rolling Stock In daily railway operations, some inevitably unexpected disruptions often occur, resulting in some temporary maintenance tasks arranged, which will perturb normal operations. This paper builds up a space-time-state network based mixed- integer linear programming model to reschedule timetable and rolling stock, i.e. retiming, reordering, re-servicing and re-circulation of rolling stocks, simultaneously. The objective function minimizes total cost of timetable and rolling stock utilization. Finally, comprehensive experiments based on Beijing - Shanghai High-speed Railway Line are conducted to examine the applicability and validity of the proposed method. n TD29 North Bldg 221B Modeling and Analysis of Emerging Mobility Services and Systems II Sponsored: TSL/Urban Transportation Sponsored Session Chair: Hai Wang, Singapore Management University, Singapore Management University, Singapore, 178902, Singapore 1 - Refuse or Accept? Analysis of Taxi Driver Operating Strategies in e-Hailing Platforms Debjit Roy, Indian Institute of Management Ahmedabad, Ahmedabad, India, Arulanantha Prabu PM, Prahalad Venkateshan In e-hailing platforms, a taxi driver adopts one of the three reactive strategies to a customer pick-up request: no refusal, refusal based on proximity, and refusal based on profitability index (based on dry and live-run distance). We develop driver operating policies for profit maximization with different service region topologies. 2 - On the Range Anxiety for Electric Vehicles: An Empirical Investigation Sang Won Kim, CUHK Business School, Hong Kong, China, Ho-Yin Mak, Marcelo Olivares, Ying Rong One of the most well-cited reasons of slow adoption of electric vehicles is the range anxiety. However, it has not been adequately quantified, quite possibly due to the lack of quality data. We propose a novel way to do so by use of a dataset from a car sharing platform. 3 - Can Multiple On-demand Service Platforms Coexist? Jiaru Bai, Binghamton University, SUNY, Binghamton, NY, 13902, United States, Christopher S. Tang As venture capital firms are financing many startups that are essentially on- demand service platforms, we wonder how many startups of this kind can survive in a competitive market. To examine this question, we present a model in which two on-demand service platforms compete in both the provider and customer markets with earning-sensitive service providers and price- and delay-sensitive customers. Rescheduling Due to Temporary Maintenance Tasks Jianrui Miao, Beijing Jiaotong University, Beijing, China

356