Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TC52

5 - Online Customer Purchase Prediction Based on Machine Learning Cheng Chen, Huazhong University of Science and Technology, Wuhan, 430074, China, Xianhao Xu Accurate prediction of online customers’ purchase behavior is of great significance to the distribution and inventory management of online retailers. Machine learning algorithm is widely concerned and applied due to its high prediction ability. In this paper, we use click stream data to predict the online customer’s purchase behavior by using five supervised learning algorithms in machine learning. Detailed result analysis validated the effectiveness of the proposed machine learning methods. 6 - Using Consumer Behavioral Data to Optimize Self-collection Points for Online Retailers Yaohan Shen, Huazhong University of Science and Technology, Luoyu Road, Wuhan, 430074, China, Xianhao Xu We proposed a procedure to optimize self-collection points for online retailers. Our research shows an optimal way to locate self-collection points in potential points and decide the number of self-collection points to trade off consumer service level and the total logistics cost for online retailers. We first cluster customers addresses, and predict the purchasing frequencies for each period from user behavioral data. Then we propose a mathematical model to optimize the Burak Cankaya, Assistant Professor, Embry-Riddle Aeronautics University, 600 S. Clyde Morris Blvd,, Daytona Beach, FL, 32114, United States, Berna Eren Tokgoz, Ali Dag This presentation proposes a machine learning based automatic labeling methodology for chemical tanker activities. The proposed methodology utilizes three machine learning algorithms to classify chemical tanker activities. The findings in this study can be used by port authorities, shipping companies, vessel operators and other stakeholders for decision support with tableau dashboard. 8 - Resource-constrained Routing With Deadlock Detection for Large-scale AGY System V Jorge Leon, Texas A&M University, College Station, TX, 77843-3367, United States, Ek Peng Chew, Loo Hay Lee A methodology that detects deadlocks while generating AGV routings is presented. The AGV transit layout is modeled as a capacitated-resource graph and used to make vehicle assignments and path decisions while generating routings. A new connected acyclic-and-transpose graph is proposed to detect a special case of cyclic-deadlocks during routing generation. The routing procedure runs in polynomial time. 9 - Exploring Dependencies Across Multiple Online Social Network Platforms Hwang Kim, Assistant Professor, Chinese University of Hong Kong, 11F Cheng Yu Tung Building, # 12, Room 1134, Shatin, Hong Kong, Vithala R. Rao Users’ interplay among these various social networking platforms implies that several sources may induce interrelationships among the platforms. To understand this, we propose an integrated visit model that accommodates networking activities across social network platforms and test the model using data from two social network gaming platforms. The model discovers a new source of dependencies that stems from communications with friends overlapping in different network platforms. The simulation study provides the managerial implications of how firms manage networks by discovering that the proportions of overlapping friends cause asymmetric spillover effects across network platforms. 10 - Pricing vs Targeting in a Reward-based Crowdfunding lei xu, Prof., Tianjin University of Technology, Tianjin, China Abstract not available. 11 - Real-time Prediction of the Optimal Con?guration of an Intelligent Gate-line System Using Queueing Theory Eric Enkele Longomo, University of Portsmouth, Yorke Street, Southsea, Portsmouth, PO5 4EL, United Kingdom, Djamila Ouelhaj, Mark Dyne This research paper builds on existing knowledge in the ?eld of Queueing Theory and proposes a mathematical model that considers a Discrete Event Simulation (DES) of passengers arriving or/and in the process of exiting a train station. The mathematical model is implemented in order to compute in real-time the optimal con?guration of a self-recon?guring intelligent gateline system at an actual overground or underground train station. The DES used in this paper, models the operation of the intelligent gateline as a discrete sequence of events in time. Each event is assumed to occur at a particular instant in time and marks a potential change of states -number of passengers and gateline con?guration, in the system. Overhead sensors were used to record passengers’ timestamps, coordinates, average speeds and arrival at the gateline in both directions of the ?ow in real- time. Maximising individual gate throughput and the average gateline throughput, walkways utilisation, cutting the overall passenger waiting time, safely managing passengers’ ?ow to prevent platform overcrowding (and potential station closure), increasing station capacity, reducing queueing in both arrival and departure and relieving station sta?s from manually operating the SCU (Station location of self-collection points, based on customers’ information. 7 - Vessel Movement Labeling with Machine Learning and Dashboard Implication

substantially. 4 - Scheduling Meter Readings for Utility Companies in Turkey Emre Eryigit, PhD Candidate, University of Massachusetts Amherst, 336 East Hadley Road, Amherst, MA, 01002, United States, Ahmed Ghoniem We examine a meter reading scheduling problem that impacts the quality of the power consumption forecast made by electricity distribution companies in Turkey. The problem is modeled as a 0-1 integer program with constraints related to manpower capacity and other industry restrictions. Exact and heuristic solutions are reported in our computational study for large-scale instances involving several thousand meter groups. 5 - Scheduling Meter Readings for Utility Companies in Turkey Emre Eryigit, PhD Candidate, University of Massachusetts- Amherst, Amherst, MA, USA, Ahmed Ghoniem We examine a meter reading scheduling problem that impacts the quality of the power consumption forecast made by electricity distribution companies in Turkey. The problem is modeled as a 0-1 integer program with constraints related to manpower capacity and other industry restrictions. Exact and heuristic solutions are reported in our computational study for large-scale instances involving several thousand meter groups. n TC52 North Bldg 231C Flash Session I General Session Chair: INFORMS, 5521 Research Park Drive, Suite 200, Catonsville, MD, 21228, United States 1 - Multiple Floors Healthcare Facility Layout Problem Ling Gai, Shanghai University, Office 432, Managent Science, School, No. 599 Shangda Road, Shanghai, 201444 We consider a multiple attributes group decision making problem for the multi- floor healthcare facility layout problem (MHFLP). Several feasible alternatives are produced first by solving a mixed-integer programming with the objective of minimizing flow costs. These alternatives are then evaluated according to their qualitative criteria by a group of experts. Given the facts that each alternative has several qualitative attributes, and the experts have different specialties on these attributes, we propose an method to determine the proper attributes weight and the experts weight (on different attributes), if they are only partially known in advance. 2 - Credit Risk Assessment for Mortgage Lending by Artificial Intelligent Underwriter Jian-Bo Yang, Professor, Director of DCSRC, The University of Manchester, AMBS East Building, Booth Street East, Manchester, M13 9SS, United Kingdom, Swati Sachan, Dong-Ling Xu Underwriting for offering mortgage loans consists of assessment on affordability, credit, repayment, security, etc. This enables precise estimation of risk or ability of borrower to repay the loans. Prevalent decision-making approaches adopted by most lenders follow pre-established rules and utilities historical data. This paper discusses a novel belief-rule-based AI model and a probabilistic machine learning process to mimic human underwriters’ inference and decision making process for improvement of productivity and quality. A case study is provided to show its potential wide applicability. 3 - Interpretable Machine Learning Algorithm via the Belief Rule Based Expert System Dong-Ling Xu, Chair Professor in Decision Science and Systems, Manchester University, Alliance Manchester Business School, F37 MBS East, Manchester, M15 6PB, United Kingdom, Jian-Bo Yang Some most popular machine learning algorithms, such as neural network, are not interpretable. In practice, however, there is an increasing need for transparent and interpretable machine learning algorithms. Our theoretical and applied research shows that the Belief Rule Based (BRB) expert system has a powerful learning ability which is capable of approximate complex nonlinear relationship arbitrarily close. In this presentation we outline the outcomes of the theoretical research and demonstrate the learning capability and the transparent property of the BRB system using examples. 4 - The Relationship between Environmental Innovation and Environmental Productivity JiangJiang Yang, University of Science and Technology of China, HeFei City, Anhui province, China Environmental innovation has been recognized as an efficient way to alleviate environmental problem. However, how environmental innovation effect environmental productivity and the whether the impact differs from various area. We used DEA approach to investigate the relationship between environmental innovation and environmental productivity.

325