Informs 2015

207 / 552

INFORMS Philadelphia – 2015
205
78 - Using Past Scores and Regularization to Create a Winning
NFL Betting Model
Eric Webb, Graduate Student, Indiana University, 1309 E. 10th
Street, Bloomington, IN, 47405, United States of America,
Wayne Winston
Is the National Football League betting market efficient? We have devised a
profitable betting model that would win 52.9% of the 7,554 bets against the
spread it would have made over 33 seasons. Scores from previous weeks are used
to estimate the point value of each team’s offense and defense. These values
predict next week’s scores, and a bet is placed against the advertised spread. The
sum of squares of offensive/defensive point values are constrained to be less than
a regularization constant.
79 - Self-organized Deliberative Agent and its Application in Medical
Claim Editing
Jack Xue, Exec. Application Architect, National Government
Services, Anthem, 8115 Knue St., Indianapolis, IN, 46250,
United States of America,
In this system each agent self-adjusts its organization per environments before
execution and optimizes itself both in structure and in execution steps to meet
Service Level Agreement. The scheduling algorithm is formatted as an LP or MILP
then generalized to stochastic with uncertainty in phase transitions. The efficacy is
demonstrated in a medical claim editing system that identifies irregularities in
million claims with calculations of terabyte current and historical data, in near
real-time.
80 - Behavioral Analysis of Participants in Community Outreach
Intervention Projects
Haoxiang Yang, Northwestern University, 2145 Sheridan Road,
Room C151, Evanston, IL, 60208, United States of America,
, David Morton,
Alexander Gutfraind
The Community Outreach Intervention Projects (COIP) serves the Chicago
metropolitan area, providing support for drug users to help prevent infectious
diseases. Using about 10 years of data, we study the behavior of participants in
COIP’s syringe exchange program, focusing on the temporal process governing
their visits to storefronts and demographics. With a better understanding of the
participants’ behavior, we aim to help develop an improved marketing plan for
COIP.
81 - A Schatten-p Norm Perturbation Inequality and its Application in
Low Rank Matrix Recovery
Man Chung Yue, The Chinese University of Hong Kong, RM
2511,Man Tai House, Tsz Man Est,, Tsz Wan Shan, KLN,
Hong Kong, Hong Kong - PRC, ,
Anthony Man-cho So
Low-rank matrix recovery, with its applications in finance, network localization,
etc, has recently attracted intense research and can be formulated as a rank
minimization. Because of the NP-hardness, a common heuristic is to use the
Schatten-p norm minimization as a surrogate. However, the equivalence property
of this remains elusive and hinges on a conjectured matrix inequality. We prove
this conjecture and derive sufficient conditions for low-rank matrix recovery
using Schatten-p heuristics.
82 - Demand Prediction and Two-stage Inventory Policy for an Online
Flash Sale Retailer
Mengzhenyu Zhang, University of Michigan, Stephen M. Ross
School of Business, 701 Tappan Ave, Ann Arbor, MI, 48109,
United States of America,
We show the work cooperated with an online flash sale retailer in China. With
millions history sales records, we use machine learning techniques to predict
demand and propose a two stage inventory policy, which requires to response
quickly to early-hour real sales and restock inventory thereafter. A model is built
to explain positive and negative effects of our policy. Experimenting on real data,
we help this firm increase profit by approximately 18% and reduce remaining
inventory by over 50%.
83 - Heuristics for Bicycle Sharing System Repositioning Problem
Mary Kurz, Clemson University, 110 Freeman Hall, Clemson, SC,
29634, United States of America, Ling Zu
This paper studies the static bicycle repositioning problem with real NYC Citi
system data. It selects a subset of stations to visit, sequences them, and determines
the pick-up/drop off quantities in each visited station. The study incorporates real
problem characteristics by minimizing total penalties of lacking/ overflowing
bicycles and routing cost. A Variable Neighborhood Search heuristic is introduced
to solve the described problem.
84 - Ranking Universities: Practices, Problems and Way Forward
Muhammad Mukhtar,Professor, American University of Ras Al
Khaimah, (AURAK), Ras Al Khaimah 10021, United Arab
Emirates, , Sarah Mukhtar,
Zahida Parveen, Brian Wigdahl
We report here a comparison of various global ranking systems of universities
and their impacts in the society. Five global ranking systems parameters evalua-
tions revealed that Times Higher Education Ranking System is more appropriate
when compared with other ranking systems. Our analyses revealed that disci-
pline wise ranking by various global rankers are creating dilemmas for the par-
ents and public to decide about their children education. We propose normaliza-
tion of ranking systems.
Monday, 1:30pm - 3:00pm
MC01
01-Room 301, Marriott
Military O.R. and Applications V
Sponsor: Military Applications
Sponsored Session
Chair: Michael Hirsch, ISEA TEK, 620 N. Wymore Rd., Ste. 260,
Maitland, FL, 32751, United States of America,
1 - Predicting the Use of Violence using Machine Learning Methods
Erkam Guresen, KHO, Dikmen, Ankara, Turkey,
Salih Tutun, Gulgun Kayakutlu
Use of Violence by Ethno–Political Organizations is threatening not only
individually countries but also all humanity. As a consequence governments are
obliged to take measures in their budget for this threat. Obviously it does not
mean that whole of security budgets consist of spending for Use of Violence,
however it has important effects on them. For all these reasons, the aim of this
study is to examine the predetermine models for use of violence.
2 - Unmanned Aerial Vehicle Routing in the Presence of Threats
Kamil Alotaibi, Taibah University, College of Engineering, P.O.
Box 344, Almadinah Almunawwarah, PC41411, Saudi Arabia,
Jay Rosenberger,
Siriwat Visoldilokpun, Stephen Mattingly
We study the routing of Unmanned Aerial Vehicles (UAVs) in the presence of
enemy threats. We formulate a mixed integer linear program that maximizes the
total number of visited targets for multiple UAVs while maintaining both the
route travel time and the total threat level to predetermined constant parameters.
Several waypoint generation methods are proposed. Branch and price is used to
solve the problem. A computational study is done and results for different
scenarios are presented.
3 - Variants of the Target Visitation Problem
Michael Hirsch, ISEA TEK, 620 N. Wymore Rd., Ste. 260,
Maitland, FL, 32751, United States of America,
In this research, we consider the target visitation problem, and discuss some
variants. Mathematical formulations are derived, heuristics are developed, and
results are presented.
MC02
02-Room 302, Marriott
Logistics and Transportation Security
Cluster: Homeland Security
Invited Session
Chair: Gary Gaukler, Drucker School of Management, Claremont
Graduate University, Claremont, CA, 91711, United States of America,
1 - Cyber Vulnerability Models
Murat Karatas, The University of Texas at Austin, 1 University
Station Austin TX 78712, United States of America,
, Nedialko Dimitrov
Infrastructures, such as university nuclear reactors, are controlled through cyber-
physical systems. Assessing the vulnerability of these system is key in directing
defensive investment. We present an MDP to compute an optimal attack policy.
The MDP has an exponential number of states, and is based on tracking the set of
available attacks for each link in the network. Surprisingly, we show it is possible
to compute values for each MDP state, and optimal attack policies, using s-t
reliability.
MC02