2016 INFORMS Annual Meeting Program

MA49

INFORMS Nashville – 2016

MA49

2 - Optimizing Ballistic Imaging Operations Can Wang, Stanford University, Stanford, CA, United States, canw@stanford.edu, Mardy Beggs-Cassin, Lawrence M Wein Ballistic imaging can solve crimes by comparing images of cartridge casings to a database of images from past crimes. Many cities lack the capacity to process all of their images. Using data from Stockton, CA, we allocate limited capacity to maximize the hit rate. The hit rate can be doubled by giving crime scene evidence priority over test fires, and ranking cartridge types by their hit rate and processing evidence from only top-ranking cartridge types. 3 - Using Informed Heuristics For Pretrial Release Jongbin Jung, Stanford University, jongbin@stanford.edu We present a simple and intuitive strategy for creating statistically informed decision rules that are easy to apply and easy to understand, in the context of pretrial release. These simple informed heuristics take the form of a weighted check list and can be applied without the aid of a computer, but perform on par with state-of-the art machine learning methods. The rules can be readily constructed with moderate statistics knowledge using common and freely available software packages, facilitating adoption by practitioners in a wide array of fields. 4 - Assessing Risk-based Policies For Pretrial Release And Split Sentencing In Los Angeles County Jails Lawrence Wein, Stanford University, lwein@stanford.edu Mericcan Usta Court-mandated downsizing of the CA prison system has led to overcrowding in CA jails. We model the flow of individuals in the Los Angeles County jail system, from arraignment through post-sentence supervision. We optimize joint pretrial release and split-sentencing policies that are based on the type of criminal charge and the risk category as determined by the CA Static Risk Assessment tool. Policies that offer split sentences to all low-level felons optimize the trade-off between public safety and jail congestion. 214-MCC Recent Developments in Humanitarian Logistics Sponsored: Public Sector OR Sponsored Session Chair: Kezban Sokat, Northwestern University, 2145 Sheridan Road, Room C210, Evanston, IL, 60208, United States, kezban.yagcisokat@u.northwestern.edu 1 - The Vaccination Campaign Routing Problem Melih Celik, Middle East Technical University, Ankara, Turkey, cmelih@metu.edu.tr, Bahar Cavdar, Haldun Sural This study considers the routing of vaccination campaigns in developing country settings, where a team selects from a set of regions to visit and sequences these visits, subject to special time window constraints. The objective is to maximize the total number of people reached. A two-stage heuristic is proposed, where the first stage solves a b-matching problem to determine the regions to visit each day, whereas the second stage solves a modified orienteering problem to determine the routes. 2 - Cash, Vouchers Or In-kind Aid: A Game Theory Approach To Determine Optimal Aid Transfers Christos Bitos, Kühne Logistics University, Hamburg, Germany, Christos.Bitos@the-klu.org, Maria Besiou This research aims to determine the conditions for optimal aid transfers in the aftermath of a disaster or during a long-term development program. By using the principles of game theory, we consider the strategic interactions between Humanitarian Organizations (HO) and local markets, and discern how these interactions affect the distribution of aid to communities. We consider the effects of local market supply chain fluctuations, and how the market fluctuates in response to local and national crises. Ultimately, we hope to develop a framework to contextualize the intricacies of humanitarian relief distribution that can be applied broadly. 3 - An Agent-based Modeling Approach To Assess Coordination Among Humanitarian Relief Providers Jessica Heier Stamm, Kansas State University, Manhattan, KS, United States, jlhs@k-state.edu, Megan Menth Coordination between humanitarian organizations during disaster response may improve efficiency, reduce duplication of efforts, and lead to better outcomes for beneficiaries. We employ agent-based simulation to examine coordination strategies among humanitarian organizations that make post-disaster location decisions regarding temporary service facilities. For example, over 4,000 temporary learning facilities were needed after the 2015 Nepal earthquakes damaged or destroyed numerous school buildings. Our model is informed by data from the Nepal response and a survey of humanitarian professionals. We find that coordination strategies impact efficiency, effectiveness, and equity. MA52

211-MCC Tutorials and Examples of Software and Methods for Social Media Analytics Invited: Social Media Analytics Invited Session Chair: Theodore T Allen, Ohio State University, Columbus, OH, 43210, United States, allen.515@osu.edu 1 - NLP, LDA, SMERT, k-Means And Efficient Estimation Methods Zhenhuan Sui, Ohio State University, sui.19@osu.edu We describe some of our recent advances in more efficient estimation for text- based clustering and topic discovery. Also, we illustrate the capability of VBA code for text processing and benchmark popular methods including k-means clustering and Latent Dirichlet Allocation in terms of computational efficiency and accuracy. 2 - Innovative Scheduling And Kriging-Based Optimization Methods In VBA Sayak Roychowdhury, Ohio State University, rowchowdhury.6@osu.edu A suite of techniques for scheduling and simulation optimization is described based on VBA and our own innovations. Results showing improved performance compared with alternatives for scheduling and inventory policy-making are described. Possible roles for the methods support social media analytics are also described. 3 - Cybersecurity Using Interdiction Murat Karatas, University of Texas, Manor Road, Austin, TX, 78722, United States, mkaratas@utexas.edu, Nedialko Dimitrov Recent cyber-attacks on private and public groups highlight the importance of a proper cybersecurity structure. Having well-structured cybersecurity decreases the vulnerability of the system. We present a network interdiction model to find the optimal strategy for a cyber physical network. Our model considers the specifics of the network structure. 212-MCC Life After the PhD: Early Career Development Panel Sponsored: Minority Issues Sponsored Session Moderator: Julie Ivy, North Carolina State University, 111 Lampe Drive, Raleigh, NC, 27695, United States, jsivy@ncsu.edu 1 - Life after The PhD: Early Career Development Panel Julie Ivy, North Carolina State University, jsivy@ncsu.edu 213-MCC New Models in Criminology Sponsored: Public Sector OR Sponsored Session Chair: Lawrence Wein, Stanford University, 655 Knight Way, Stanford, CA, 94305, United States, lwein@stanford.edu 1 - Machine Learning For Crime Series Detection And Criminal Recidivism Prediction Cynthia Rudin, Duke University, LSRC Building D101, 308 Research Drive Campus Box 90129, Durham, NC, 27708-0129, United States, cynthia@cs.duke.edu, Berk Ustun, Tong Wang I will discuss machine learning algorithms for two problems: crime series detection, and recidivism prediction. In crime series detection, the goal is to identify crimes that were committed by the same individual(s). We cast this as a clustering problem with cluster-specific feature selection, in joint work with the Cambridge Police Department. The recidivism prediction problem is cast as a supervised classification problem, where the goal is to produce a scoring system, which is a sparse linear model with integer coefficients. This work was a finalist in the 2015 Doing Good with OR competition, and part of the winning entry of the 2016 Innovative Applications in Analytics Award. MA51 MA50

138