Informs Annual Meeting 2017

TD65

INFORMS Houston – 2017

4 - Automated Learning of Geometric Shape Deformation Models in Additive Manufacturing Raquel Ferreira, Purdue University, West Lafayette, IN, 47907, United States, rdesouz@purdue.edu, Arman Sabbaghi, Qiang Huang A major challenge in quality control for 3D printing is the specification of geometric shape deformation models. The current practice of constructing tailor- made deformation models for each combination of computer-aided design model, 3D printer, material, and machine setting, is impractical for general application. We propose an automated model building method based on extreme learning machines that does not require detailed knowledge of the underlying system. We illustrate promising benefits of our methodology for real-life case studies. Comparisons with current tailor-made models demonstrate the potential of our automated approach to yield similar results in a more systematic manner. 370D Transportation, Public Contributed Session Chair: Nishanth Mundru, Massachusetts Institute of Technology, Operations Research Center, E40, Cambridge, MA, 02139, Siyang Xie, U. of Illinois at Urbana-Champaign, 205 N Mathews Avenue, 3150 Newmark Lab, Urbana, IL, 61801, United States, sxie13@illinois.edu, Yanfeng Ouyang Traditional districting problems focus on partitioning an area into districts under operational criteria, e.g., contiguity, compactness, workload balance, etc. In many real-world applications, each district is associated with a facility, thus involving additional location and assignment decisions. We develop a facility districting framework to combine the facility and districting considerations, with possible facility disruptions. We formulate the problem as a set-covering based reliable network partitioning model and design a column generation algorithm to effectively solve it. Numerical case studies are conducted to demonstrate the framework and to draw managerial insights. 2 - Electric Vehicle Routing Problem under the Limited Energy Supply In recent years, the development and diffusion of Electric Vehicles have been advanced. For the delivery companies, electric vehicles are expected not only for means of transportation and delivery but also energy management systems by using vehicle storage batteries. Particularly, in cooperation with the smart grid or virtual power plant, which has recently undergone reforms, it is expected to minimize energy usage and reduce electricity bills. In this research, we will develop a delivery plan by electric vehicles according to changes in energy supply in future electric power systems. 3 - A Modern Optimization Approach for the Robust Airlift Planning Problem for the United States Transportation Command Nishanth Mundru, Massachusetts Institute of Technology, Operations Research Center, E40, Cambridge, MA, 02139, United States, nmundru@mit.edu, Dimitris Bertsimas, Allison An Chang, Velibor Misic The United States Transportation Command (USTRANSCOM) plans missions globally, the majority traveling by air. These missions are challenging to plan and are subject to uncertainty in travel times and cargo weights. We propose a solution approach based on local search methods, robust and mixed-integer optimization, and column generation, and show that it provides high quality solutions in practical times. 370E Supervised Learning Models and Applications Sponsored: Data Mining Sponsored Session Chair: Ali Dag, Ali.Dag@usd.edu 1 - A Supervised Learning Approach for Matched Case-control Study Nooshin Shomal Zadeh, Arizona State University, Tempe, AZ, United States, nshomalz@asu.edu, Sangdi Lin, George Runger Matched case-control study is commonly used in the field of public health to detect important variables that affect an outcome. One of the challenges in this study is to determine the relationship between the variables and the outcome in the presence of large number of variables whose interactions can affect the outcome. Another challenge is that the relationships can be nonlinear and some Shinichi Maeda, Osaka University, Suita, Japan, shinichi.maeda@ist.osaka-u.ac.jp, Hiroshi Morita TD64 United States, nmundru@mit.edu 1 - Reliable Spatial Districting TD63

are difficult to model by traditional approaches. In addition, handling both numerical and categorical variables remains an important and yet difficult problem. In this research, we propose a supervised learning approach which can address these challenges and handle both categorical and numerical variables. 2 - A Novel End-to-end Deep Learning Model for Time Series Classification Sangdi Lin, Arizona State University, AZ, United States, Sangdi.Lin@asu.edu, George Runger Time series classification is an important task in many domains such as manufacturing and finance. It is gaining more attention in health care as the wearable devices become popular, and more health-related time series data is recorded. In this presentation, we introduce a novel deep learning model which can be generally applied to time series classification tasks in various domains. The proposed deep learning model can be trained in an end-to-end fashion without the need of the traditional exhausting feature engineering step. Our deep learning model not only produces accurate classification results, but also obtains some insights and understanding about the time series data. 3 - Predicting the Survival of Breast Cancer Patients via Supervised Machine Learning Algorithms Eyyub Kibis, PhD, The College of Saint Rose, Albany, NY, United States, eyyubyunus@gmail.com, Esra Buyuktahtakin, Ali Dag Breast cancer is the second leading cause of death after lung cancer in the U.S. There are many reasons that impact the life time expectancy of patients. In this study, given a set of explanatory variables that include demographics, health conditions, and cancer treatment regimens of patients, our objective is to determine patient specific survival scores ten years after the initial breast cancer diagnosis. We use data mining methods and information fusion technique to determine the key features for breast cancer survivability as well as the patient specific survival score. Our results can help practitioners for adjusting the aggressiveness of treatments based on the survival scores of patients. 370F Applications in Freight Transportation and Logistics Sponsored: TSL, Freight Transportation & Logistics Sponsored Session Chair: Hakan Yildiz, Michigan State University, Department of Supply Chain Management, 632 Bogue St., East Lansing, MI, 48824, United States, yildiz@msu.edu 1 - Reliable Synchromodal Transportation Networks Rob A. Zuidwijk, Professor, Ports in Global Networks, Erasmus University-Rotterdam, RSM.Erasmus University, P.O. Box 1738, Rotterdam, 3000 DR, Netherlands, rzuidwijk@rsm.nl This paper studies planning of container transport services on port hinterland networks where various transport resources are deployed in a flexible way. Two aspects receive particular attention. First, the sourcing of transport capacity to provide these services. Second, the reliability of the door-to-door services based on the reliability of the scheduled services of the various modes. 2 - Drone-assisted Delivery: A Practical Approach Iman Dayarian, Georgia Institute of Technology, School of Industrial and Systems Engineering, 765 Ferst Dr NW, Atlanta, GA, 30318, United States, iman.dayarian@isye.gatech.edu, Martin W. P.Savelsbergh Unmanned aerial vehicles, drones in particular, have recently seen an increased level of interest as their potential use in same-day home delivery has been promoted by large retailers and courier companies. We introduce the vehicle routing problem with drone resupply in which a delivery vehicle makes deliveries in a certain area and is regularly resupplied by a drone. Resupply can take place at specific meeting locations, where the drone will land on the roof of the delivery vehicle. We explore and compare different decision-making strategies and provide TD65 Hakan Yildiz, Michigan State University, Department of Supply Chain Management, 632 Bogue St., East Lansing, MI, 48824, United States, yildiz@msu.edu, Robert Wiedmer Empty beverage containers are mostly returned to retail stores, which are then picked up and transported to processing facilities. The routes and schedules of pickup vehicles are mostly done based on forecasts, which makes this problem a variant of Vehicle Routing Problem with Stochastic Demand. We analyze data and develop methods to optimize the pick-up and processing operations. insights resulting from a comprehensive computational study. 3 - Transportation and Process Planning for Beverage Container Recycling

399