Informs Annual Meeting 2017

TB67

INFORMS Houston – 2017

2 - Examining Multi-classification Methods for Consumer Choice Matthew A. Lanham, Purdue University, West Lafayette, IN, United States, lanhamm@purdue.edu This study examines multi-classification methodologies that a retailer might employ to gauge the propensity a product might sell when purchase substitution is inevitable. This study is important because retailers are routinely reevaluating the mix of products they offer consumers and often use predictive models that enable them to estimate the substitutability of their products. We model the substitution propensity for a set of products using a national retailer’s data using common and less-common multi-classification methodologies and compare the results. We highlight where certain models perform better than others and discuss our future extensions of this study. 3 - A Two-stage Learning Approach to Improve New Item Forecasts in Retail Inderjeet Singh, Infor Inc., Atlanta, GA, 30309, United States, Inderjeet.Singh@infor.com, Ronald P. Menich, Ashwin Mishra Forecasting new items for planning, vendor ordering, pricing, allocation and replenishment functions is a challenging problem in retail. To forecast new items, legacy forecasting solutions require users to manually set up one or more clone item(s), leading to a tedious process. At Infor, we use an in-house implementation of Factorization Machines to generate more accurate forecasts. Our attribute- based demand forecasting solution offers clone free forecasts for new items. We recently implemented a simple two-stage learning approach to further improve our new item forecasts. We present real-life examples to elaborate the approach and talk about an implementation experience. 4 - Estimating Demand for Substitutable Products when Inventory Records are Unreliable Daniel Waymouth Steeneck, Air Force Institute of Technology, WPAFB, OH, United States, steeneck@mit.edu, Fredrik Eng-Larsson, Francisco J. Jauffred We present a procedure for estimating demand for subsitutable products when the inventory record is unreliable and only validated infrequently and irregularly. The procedure uses a structural model of demand and inventory progression, which is estimated using a modified version of the EM algorithm. 370F Same-Day Delivery Logistics Sponsored: TSL, Freight Transportation & Logistics Sponsored Session Chair: Ozlem Ergun, o.ergun@northeastern.edu Co-Chair: Necati Duman, duman.n@husky.neu.edu 1 - A Robust Optimization Approach for Crowdsourcing Last-mile Deliveries Soraya Fatehi, University of Washington, 5215 15th Avenue, Apt 15, Seattle, WA, 98105, United States, sfatehi@uw.edu, Michael R.Wagner We study a robust optimization model of last-mile deliveries motivated by the Amazon Flex and Uber Eats programs. In these applications, the crowd is utilized for last-mile deliveries. We derive the optimal usage of the crowd to supplement traditional delivery options, to minimize the delivery cost per package, under delivery window constraints. 2 - Peer-to-Peer Transshipment in On-demand Same Day Delivery Jane Lin, University of Illinois-Chicago, 842 W. Taylor Street (M/C 246), Chicago, IL, 60607, United States, janelin@uic.edu In light of the emerging on-demand, same-day (ODSD) delivery service market, this study proposes, formulates, and evaluates a new ODSD delivery strategy using direct peer-to-peer transshipment (P2PT). P2PT involves direct en-route package relays among multiple couriers to extend beyond the normal service range of a single courier in order to reach the package’s final destination. Our results show that P2PT improves initially with the economy of scale, and then flattens out in the static demand scenario or worsen in the real-time demand scenario. Compared with direct shipping, P2PT is shown a low cost strategy for the ODSD service with the economy of scale. 3 - Order Acceptance Mechanisms for Same-Day Delivery Mathias A. Klapp, Pontificia Universidad Catolica de Chile, Santiago, 7820436, Chile, maklapp@ing.puc.cl, Alan Erera, Alejandro Toriello We study same-day delivery by formulating the Dynamic Dispatch Waves Problem with Immediate Acceptance that models integrated request management and order distribution decision-making where delivery requests arise dynamically throughout the day. When a request arises, a decision is made immediately to accept (offer service) or reject (with a penalty). All accepted requests are included in dynamically-updated dispatch plans that serve each request by the end of the day. We provide computational experiments that estimate a cost increase of 4.4% when imposing immediate order acceptance. TB65

4 - A Study to Improve Routing Precision in Same Day Delivery Necati Oguz Duman, Northeastern University, Boston, MA, United States, duman.n@husky.neu.edu, Mehmet Talha Dulman, Ozlem Ergun Most of the studies on vehicle routing problems assume that distances between nodes are given as parameters and they are accurate. However, in most real world applications, the nodes are given and distances between each pair must be acquired either from a map provider or with an approximation method. In this study, we analyze the relationship between driving distance and straight line distance and provide a regression model to predict the driving distance. To test the efficiency of our model, we use a simple routing algorithm with clustering and insertion methods in construction and several local search methods in improvement phases. 371A TSL Special Session Sponsored: Transportation Science & Logistics Sponsored Session Chair: Lavanya Marla, U of Illinois at Urbana-Champaign, 104 S Mathews Avenue, Room 216E, Urbana, IL, 61801, United States, lavanyamarla@cmu.edu 1 - TSL Lifetime Achievement Award Winner Lavanya Marla, U of Illinois at Urbana-Champaign, 104 S. Mathews Avenue, Room 216E, Urbana, IL, 61801, United States, lavanyam@illinois.edu The TSL Lifetime Achievement Award winner, announced in the TSL Business Meeting on Monday, will present a plenary address to the TSL community. TB66

TB67

371B Tutorial: Analytics Applications for Additive Manufacturing Sponsored: Quality, Statistics and Reliability Sponsored Session

Chair: Linkan Bian, Mississippi State University, 260 McCain Building, 260 McCain Building, Mississippi State, MI, 39762, United States, bian@ise.msstate.edu Chair: Alaa Elwany, Texas A&M University, 3131 TAMU, College Station, TX, 77843-3131, United States, elwany@tamu.edu 1 - Tutorial: Analytics Applications for Additive Manufacturing Linkan Bian, Mississippi State University, Industrial and Systems Engineering Department, P.O. Box 9542, Mississippi State University, MS, 39762, United States, bian@ise.msstate.edu, Alaa Elwany Data analytics is playing a key role in advancing numerous research and application areas. Additive manufacturing is a particular area that can benefit greatly from data-driven analytics and predictive modeling, especially when properly integrated with relevant physics-based models and experimental methods. This tutorial will take a deep dive into how tools of data analytics can be applied in the modeling, analysis, and control of Additive Manufacturing processes. Topics in process planning and optimization, materials informatics, process monitoring, and quality control will be the focus of the tutorial.

327