2016 INFORMS Annual Meeting Program

WE20

INFORMS Nashville – 2016

WE18 106A-MCC DMA Machine Learning Contributed Session Chair: Hang Li, Pennsylvania State University, University Park, PA, United States, Huli80@psu.edu 1 - Risk Prediction On Life Insurance Lapse Ceni Babaoglu, Dr., Ryerson University, Ryerson University, 350 Victoria Street, Toronto, ON, M5B 2K3, Canada, cenibabaoglu@ryerson.ca, Atakan Erdem, Ayse Bener Lapse constitutes a material risk for life insurance companies and needs to be controlled and managed carefully. In this project, we study the risk prediction on life insurance lapse of an insurance company. The data that we mine includes demographics, household income, unemployment and geographical information of the clients. We build a model for the prediction of lapse by using machine learning techniques. 2 - Visualization Strategies For Prediction And Classification In Supervised Machine And Statistical Learning Alexander Engau, Associate Professor, University of Colorado Denver, Denver, CO, United States, aengau@alumni.clemson.edu, Paola Gonzalez Supervised machine and statistical learning is a key task in data mining and many areas of human decision making including finance, business and industry as well as health care, medicine and bioinformatics. To facilitate a better understanding of current classifiers for prediction whose performances are typically measured and compared only numerically using cross validation, here we present a novel idea for an additional and much more meaningful visualization. We also report its recent use in two financial and medical case studies for substantial new insights into several current state-of-the-art implementations of support vector machines, decision trees, boosting and discriminant analysis. 3 - An Anomaly Detection Algorithm Using Tree-based Phase Space Method Cheng-Bang Chen, Penn State University, 445 Waupelani Dr., Apt K18, State College, PA, 16801, United States, czc184@psu.edu The cost of out of control events is usually extremely high, but the anomaly patterns are sometimes hard to detect because of the nonlinear and the high dimensional signal. Current methods focus on single signal source or dimensionality reduction, but it decreases the accuracy and sensitivity. We propose an efficient method to detect the anomaly patterns in high dimensionality accurately, using the q-tree structure for phase space, and a tree structure indexing for the subsequence signals. 4 - Optimal Experimental Design On Non-euclidean Spaces For Active Learning Hang Li, Pennsylvania State University, University Park, PA, United States, Huli80@psu.edu, Enrique Del Castillo, George Runger An Active Learning (AL) strategy selects instances to label in order to improve a model with a relatively small number of queries, accelerating learning. In recent years a number of machine learning authors have noticed the similarities between AL used for linear models and the optimal experimental design problem. In this presentation we will discuss optimal experimental design for active learning in curved spaces. A double penalized least squares functions leads to a generalization of the notions of alphabetic optimality in classical optimal design. The impact of these penalization parameters on the designs are discussed. Extensions to other types of non-euclidean spaces will be discussed.

effectiveness of the proposed model and the efficiency of the proposed solution methods. 2 - A Cross Entropy Approach To The Single Row Facility Layout Problem Xiu Ning, Tsinghua University, Shunde Building, Room 519A, Beijing, 100084, China, ningx13@mails.tsinghua.edu.cn The single row facility layout problem (SRFLP) is to arrange a given number of facilities along a straight line so as to minimize the total cost associated with the interactions between the facilities. In this paper, a metaheuristic algorithm based on the cross-entropy (CE) method is developed to solve this problem. To speed up the convergence of the algorithm, we incorporated local search procedures and symmetry breaking techniques with the CE method. The proposed algorithm has been tested using the instances available in the literature. The computational results show that the proposed algorithm can find the best solutions obtained so far for instances with up to 100 facilities. 3 - Coordinated Dynamic Demand Lot Sizing And Delivery Scheduling Problem With Resource Restriction Rui Liu, PhD, Huazhong University of Science and Technology, Wuhan, 430073, China, rliuhust316@gmail.com, Lin Wang Coordinated strategy is often used to cut down cost and increase profit in supply chain management. A new coordinated dynamic demand lot sizing and delivery scheduling problem with resource restriction is proposed and formulated. The delivery policy is integrated into coordinated dynamic demand lot sizing problem with resource restriction. In fact, the proposed model is more practical. 4 - Optimal Communication Of Information For Warfighter Benefit Azar Sadeghnejad, Buffalo, 157 Ranch Trail, Williamsville, NY, 14221, United States, azarsade@buffalo.edu, Michael Hirsch, Hector Juan Ortiz-Pena There has been a significant increase in the number of sensors deployed to accomplish military missions. These sensors might be on manned or unmanned resources, and might collect quantitative and/or qualitative information important for mission success. Of critical importance for mission success is ensuring that the collected information is routed to the people/systems that need the information for the proper making of decisions. This research mathematically formulates the problem of information routing and collection on a temporally varying communication network, and discusses some heuristics for efficient solutions. 5 - Optimization Of Information Collection And Distribution Across A Limited Communications Network McKenzie Worden, CUBRC, Inc., 4455 Genesee St., Suite 106, Buffalo, NY, 14225, United States, mckenzie.worden@cubrc.org, Azar Sadeghnejad, Chase Murray, Mark Henry Karwan, Hector Juan Ortiz-Pena For this problem, we aim to develop a heuristic that defines information collection and exchange between unmanned resources and a control station. Decisions will be made determining the routes of each resource, as well as the areas from which their sensors will collect information. Considering bandwidth and communication range restrictions, the heuristic will determine when collected information will be sent back to the control station. We will also consider scenarios in which resources may send information amongst each other, as relays, prior to the ground station receiving the information.

WE20 106C-MCC Health Care, Other I Contributed Session

WE19 106B-MCC Opt, Heuristic Programming Contributed Session

Chair: Jingyun Li, Assistant Professor, California State University - Stanislaus, 1 University Circle, Turlock, CA, 95382, United States, jli9@csustan.edu 1 - The Lag In Service Encounter: Indian Healthcare Insurance Context Sudipendra Nath Roy, Fellow Program in Management, Indian Institute of Management, Indore, Prabandh Shikhar, Rau- Pithampur Road, Indore, 453331, India, f13sudipendrar@iimidr.ac.in, Bhavin J Shah, Hasmukh Gajjar Indian healthcare service providers primarily operate through third party administrator (TPA) for bill settlement for the services that are covered under a paid medical insurance cover. Patients usually have to wait for more than acceptable time for final settlement because of coordination inefficiencies between TPA and hospital administration. This study explores the potential process improvement to overcome such delays in Indian tertiary hospital setting.

Chair: McKenzie Worden, CUBRC, Inc., 4455 Genesee St., Suite 106, Buffalo, NY, 14225, United States, mckenzie.worden@cubrc.org 1 - Quay Crane Scheduling Problem With Considering Tidal Impact And Fuel Consumption Yu Shucheng, Doctor, Shanghai University, Shang Da Road 99, Shanghai 200444, China, Shanghai, 200444, China, yushucheng2007@163.com This study investigates a quay crane scheduling problem with considering the impact of tides in a port and fuel consumptions of ships. A mixed-integer nonlinear programming model is proposed. Some nonlinear parts in the model are linearized by approximation approaches. For solving the proposed model in large-scale problem instances, both a local branching based solution method and a particle swarm optimization based solution method are developed. Numerical experiments with some real-world like cases are conducted to validate the

489