Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

WD62

3 - Rush Purchasing for Hoarding during Natural Hazards – An Empirical Study in Agricultural Products Supply Chain Hsintsz Kuo, National Taiwan University, New Taipei, 24943, Taiwan, Jiuh-Biing Sheu Disruption management is the important part in the recovery process. This paper proposes a research model to investigate the causal relationships among internal influence and external influence, rush purchasing behavior and its antecedents after huge natural hazards. We argue that the emotional contagions can moderate the relationship among rush purchasing behavior and psychological response after a huge natural hazard. 4 - A Baseline Approach to Supply Chain Risk Assessment Jo π o Dias da Silva, PhD Researcher, University of Porto, Porto, 4149-002, Portugal, Alcib ades Guedes Over the past decades, global manufacturing and logistics networks have attained unprecedented levels of complexity, with supply chains becoming ever more prone to operational disruptions of unforeseen causes and consequences. This research project takes a fresh look at fundamental supply chain risk definitions and determinants, and proposes a new generic, coherent, repeatable and primarily intrinsic methodology to measure and rate the propensity for operational disruptions in supply chain systems. The rationale of the construct, some implementation examples, the relevance for theory and practice, as well as some anticipated limitations and potential improvements are discussed. 6 - Commodity Farming in Developing Countries with an Application to the Cotton Supply Chain in Mozambique Jian Li, Northeastern Illinois University, 5500 N. St Louis Avenue, Chicago, IL, 60625, United States, Maqbool Dada, Panos Kouvelis We develop a model of cotton supply chains in which two farmers, who are supported by two ginners, make two types of fundamental decisions under yield uncertainty: how much land to allocate to cotton production before the start of the growing season, and, the pricing rules followed by ginners that purchase the cotton harvest. The focus of the study is on characterizing the optimal decisions and the impact of a government-specified minimum price. n WD62 West Bldg 103A Joint Session DM/Practice Curated: Data Science in Transportation Sponsored: Data Mining Sponsored Session 1 - Studying the Efficiency Limit of Ride Sharing Mustafa Lokhandwala, PhD Candidate, Purdue University, West Lafayette, IN, 47906, United States, Hua Cai Dynamic ride sharing offers passengers a facility to share a ride with others with the flexibility of choosing their own pick up and drop off locations while increasing the occupancy of vehicles up to a certain limit. We study this efficiency limit of ride sharing in autonomous electric vehicles using an agent-based model. Preliminary results have indicated that during peek hours, about 30% of the vehicles are available to accept new shares, but an approximately equal number of vehicles do not accept new shares. The most common reason (about 80%) that a taxi does not search for new shares has been found to be the tolerance limit of the rider to deviate from his route. 2 - Dynamic Origin Destination Estimation Based on Time Delay Correlation Among Zonal Location-based Social Network Check-in Data Jing Jin, Assistant Professor, Rutgers, The State University of New Jersey, CORE 613, 96 Frelinghuysen Rd., Piscataway, NJ, 08854, United States, Wangsu Hu Location-based Social Network (LBSN) services allow users to confirm their current locations by “check-in with places of interests. Through analyzing the check-in arrival sequences, LBSN data can identify and quantify the spatial- temporal correlation of travel demand between pairwise zones to generate the dynamic Origin-Destination (OD) flow matrix. Empirical insights can be obtained by analyzing the predicted dynamic OD flow patterns for different land use types, day-of-week, and time-of-day periods.

n WD63 West Bldg 103B Data Science for Health, Environment, and Energy Sponsored: Data Mining Sponsored Session Chair: Yasin Unlu, LLamasoft Inc, Ann Arbor, MI, 48105, United States 1 - Algorithms for the Upper Bound Mean Waiting Time in the Single-server Queue Yan Chen, Columbia University, New York, NY, 10027, United States, Ward Whitt Effective algorithms are developed to compute the tight upper bound of the mean steady-state waiting time in the GI/GI/1 queue given the first two moments of the interarrival-time and service-time distributions. A key step is to exploit the Marshall (1968) representation of the mean waiting time in terms of the idle-time distribution, which is insensitive to the rare event of the large service time. The algorithms are aided by reductions of these special queues to D/GI/1 and GI/D/1 models. One numerical algorithm exploits a negative binomial recursive formula, while another exploits a discrete-time Markov chain recursion. The computational efficiency for different methods is compared. 2 - Baum Welch Method with Limiting Distribution Constraints Daniel Waymouth Steeneck, Air Force Institute of Technology, Wright-Patterson AFB, OH, United States, Fredrik Eng Larsson Training hidden Markov models (HMMs) is traditionally done using the Baum- Welch algorithm. However, in many cases, limiting distribution information is known about the hidden states via infrequent and expensive tests. Exploiting this information, we show how to modify the Baum-Welch algorithm to obtain a more accurate estimate for an HMM’s transition matrix. 3 - Robust Estimation of Controlled Hawkes Process Michael Mark, PhD Candidate, EPFL, Route Cantonale, Lausanne, 1015, Switzerland, Thomas A. Weber The identification of Hawkes-like processes can pose significant challenges. This applies even in a linear setting, as in an established model for the repayment arrivals of delinquent credit-card accounts where a bank controls the payment- arrival intensity using suitable account-treatment actions. Despite substantial amounts of data, standard estimation methods show significant bias or entirely fail to converge. To overcome these issues, we propose an alternative method based on expectation-maximization algorithm, that exploits the natural internal branching-structure of the process, thus providing additional information to the estimator. 4 - Unsupervised Modeling and its Effectiveness for Fault Detection in Lacking Labeled Data Hyunseop Park, PhD Candidate, Pohang University of Science and Technology, 77 Cheongam-Ro, Pohang, 37673, Korea, Republic of, Junhyuk Choi, Youngho Cho, Hyunbo Cho, JaeHyuk Kim Fault detection in advance can reduce production losses due to equipment failure and unplanned stoppage. However, development of a supervised prediction model to detect faults is hindered by a lack of labeled data. Here, we present several unsupervised modeling methods for fault detection in cases that lack labeled data. The accuracy and computational time of the proposed modeling methods will be detailed and illustrated for various application conditions. Experimental results show that the unsupervised modeling methods yield Yasin Unlu, Senior Operations Research Scientist, LLamasoft Inc, McKinley Towne Center, Ann Arbor, MI, 48104, United States A typical product usually goes through different stages during its existence in the market; namely, launch (born), growth, maturity, decline and obsolescence. The duration of each stage is often of interest for making short or longer managerial decisions in an organization. We use s-shape growth models to predict the duration of each stage; namely, Logistics, Modified Logistic, Richards, Gompertz, Bass and Monomolecular as well as a tournament of these 6 models to provide the best modeling option. In order to improve the accuracy of predictions, we present different strategies for the parameter estimations of these models. significant accuracy when application conditions are considered. 5 - Product Stage Analysis with Growth Curve Models

514