2016 INFORMS Annual Meeting Program

WC03

INFORMS Nashville – 2016

WC02 101B-MCC Data Mining Applications in Health Care Sponsored: Data Mining Sponsored Session Chair: Eric Swenson, US Army, 643 Belmont Circle, State College, PA, 16803, United States, eswen75@gmail.com 1 - Identification Of Flu Hubs Using A Scale Free Network Of Flu Distance Hootan Kamran, PhD Candidate, University of Toronto, Department of Mechanical and Industrial Engin, Room RS 311, Toronto, ON, M5S 3G8, Canada, hootan@mie.utoronto.ca, Michael W Carter, Dionne Aleman, Kiearn Moore Influenza is among the leading causes of death in the world. Rapid changes in influenza virus make permanent immunity through vaccination an unviable solution, and signify the importance of surveillance systems. Current systems aggregate data based on predefined geopolitical divisions, and neglect historically significant inter-regional time connections. We have devised a network structure to model historic inter-regional flu distances in Ontario. We show that the resulting network is not a random network and in fact, exhibits behaviours of a scale-free network. The scale-free property helps us identify highly-connected regions as flu hubs, which can be prioritized in containment policies. 2 - Children Segmentation Based On Risk Of Chronic Diseases Nooshin Hamidian, Resaerch Assistant, University of Tennessee at Knoxville, 301 Woodlawn Pike, Apt A5, Knoxville, TN, 37920, United States, nhamidia@vols.utk.edu, Jafar Namdar, Rapinder Sawhney Type 2 diabetes and obesity has increased among children during the last 3 decades. The main purpose of this study is to provide a framework that identifies children who are at risk of diabetes and obesity. We explore a group of demographic and behavioral characteristics, which increase the chance of these diseases. Once the risk factors have been determined we develop a preventive model. This model determines who is at risk of these diseases. Preventing chronic diseases not only is beneficial for patients and their families, but also from the hospital point of view, it can be a solution for cutting cost and increasing hospitals’ revenue. 3 - Health Market Segmentation And Classification Of Total Joint Replacement Surgery Patients Eric Swenson, PhD Student, US Army, 643 Belmont Circle, State College, PA, 16803, United States, ers187@psu.edu Eric Swenson, PhD Student, Pennsylvania State University, University Park, PA, 16802, United States, ers187@psu.edu, Nathaniel Bastian, Harriet Black Nembhard Understanding healthcare consumers’ behaviors and attitudes is critical information when it comes to delivering patient-centered care. We apply a two- stage methodology using supervised and unsupervised machine learning methods to a 21 month sample of total joint replacement patient data. Patients cluster into 6 distinct market segments from which the cluster assignment is used as the response variable in supervised learning to classify patients. The classification model accurately predicts the cluster assignment for out-of-sample patients, while offering insight into patient behaviors and attributes to help clinicians, health marketers, and consumers enhance patient-centered care. WC03 101C-MCC Big Data III Contributed Session Chair: Mahamaya Mohanty, Research Scholar, IITDelhi, Shaheed Jeet Singh Marg, New Delhi, 110016, India, mahamayamohanty@gmail.com 1 - Establishing A Big Data Analysis Framework For Computing Nash Equilibrium With Vehicle Data Lee Yu-Ching, National Tsing Hua University, Hsinchu, Taiwan, yclee@ie.nthu.edu.tw, Ciou Si-Jheng, Huang Yi-Hao This paper provides scalable framework to handle the data unable to be dealt with by the general software. The aim is to generate a quantifiable value to represent the customers’ willingness to buy products. Finally, the proposed method is further validated by the real data of vehicles.

4 - Tractable Approximations Of Distributionally Robust Chance Constraints In Radiation Therapy Azin Khabazian, Research Assistance, University of Houston, 5465 Braesvalley Dr. Apt 566, Houston, TX, 77096, United States, akhabazian@uh.edu, Maryam Zaghian, Gino J. Lim Quadratic approximations of the distributionally robust chance constraints are developed for treatment planning to guarantee the probabilistic constraint when only partial information of the random dose contribution is known. Robust chance constraints can be conservatively approximated by second-order cone programming. In this study, we explore the condition in which the constraints depend quadratically on the random parameter, and develop more precise approximations for robust chance constraints. We evaluate these approximations in the context of a radiation therapy treatment planning problem and numerically demonstrate its superiority over the affine assumption of the constraints. 5 - Hospitalist’s Service Mix And Impacts On Length Of Stay Masoud Kamalahmadi, Doctoral Student, Indiana University, 1309 E Tenth St, Bloomington, IN, 47405, United States, maskamal@iu.edu, Kurt Bretthauer, Alex Mills, Jonathan Helm Hospitalists are physicians that specialize in caring for hospital inpatients, replacing a primary care physician who may only make rounds once per day and thereby reducing delays. Given a limited number of hospitalists in a hospital, we seek to determine their optimal service mix (workload and patient types).

Wednesday, 12:45PM - 2:15PM

WC01 101A-MCC Data Mining in Aviation Sponsored: Data Mining Sponsored Session

Chair: Nima Safaei, Bombardier Aerospace., Unit 701, 23 Rean Dr., Toronto, ON, M2K 0A5, Canada, nima.safaei@aero.bombardier.com 1 - Multivariate Analysis Of Flight En Route Efficiency Yulin Liu, University of California - Berkeley, 107 McLaughlin Hall, Berkeley, CA, 94709, United States, liuyulin101@berkeley.edu, Michael O Ball, Mark M Hansen We apply clustering and regression techniques to a large flight level trajectory dataset that includes associated traffic management initiative (TMI) data. The results quantify how TMIs, weather and other factors impact en route flight efficiency. We further evaluate the variations of en route efficiency across city pairs and over time. 2 - Data Clustering Using A Network Flow Problem To Study The Aircraft Component Failure Nima Safaei, Senior Specialist, Bombardier Aerospace., Unit 701, 23 Rean Dr., Toronto, ON, M2K 0A5, Canada, nima.safaei@aero.bombardier.com An integer programing model based on the network flow problem is proposed to cluster the categorical variables and their attributes. The variables are related to the age-related and -unrelated factors affecting the aircraft component failure. The proposed model split the variables’ attributes into a number of clusters with maximum transitive dependencies within each cluster 3 - Improving Airline Fuel Burn Predictions Using Super Learner Lei Kang, Graduate Student Researcher, University of California- Berkeley, 107 McLaughlin Hall, Berkeley, CA, 94720, United States, lkang119@gmail.com, Yulin Liu, Mark M Hansen Accurate flight fuel burn predictions are crucial in the aviation industry. By levering a large flight level fuel consumption dataset provided by a major US airline, we propose to use integrated LASSO selection and Adaboost algorithm to combine various machine learning algorithms into a super learner which can help significantly reduce the fuel burn prediction error compared to our study airlines flight planning system. The potential benefit of improved fuel burn predictions will be quantified in terms of fuel savings.

427