Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

MD61

3 - Consistent Nursing Staff Planning Under Heterogeneous Service Demand Nan Kong, Purdue University, 206 South Marting Jischke Drive, Biomedical Engineering, West Lafayette, IN, 47906-2032, United States, Mingyang Li, Mingzheng Wang The aging population in the U.S. demands a significant amount of long-term care (LTC) services. As major LTC service providers, nursing homes (NHs) are responsible for caring the frailest population of elderly adults with 24/7 formal assistance for a variety of services and supports. To improve the quality of care and realize resident-entered NH care, we develop accurate predictive models for characterizing complex service demand of individual NH residents. We also develop more proactive and cost-effective NH staff optimal planning model with a stochastic programming approach. 4 - Integrating New Testing Mechanisms Into Emergency Department Workflow Jonathan Helm, Arizona State University, W.P. Carey School of Business, 300 E. Lemon St, Tempe, AZ, 85287, United States, Zhongjie Ma, Pengyi Shi, H. Sebastian Heese In medical research, new diagnostic tests are developed and evaluated solely on their efficacy in detecting an illness. However, ignoring the workload impact of introducing new tests into existing workflow can create barriers to adoption, particularly in busy units e.g. Emergency Department (ED). In collaboration with an ED physician, we develop a framework for evaluating the workload impact of adopting new tests to bridge the gap between medical research and clinical workflow. Our queueing framework is applied to a new test, D-dimer, for detecting Pulmonary Embolism in EDs to understand what characteristics make adopting a test feasible and how to best integrate the D-dimer into the ED workflow. 5 - An Optimization Model for Collecting and Distributing Blood Products Ayca Erdogan, San Jose State University, Davidson College of Engineering, Industrial and Systems Engineering, San Jose, CA, 95192, United States, Nazanin Nader We present an integer programming model to find the optimal routes for a blood donation center. The model minimizes the total travel time for vehicles that collect donated blood from donation locations < distribute processed blood to several hospitals, considering lateness thresholds in order to minimize the blood wastage. Telecommunications Optimization Modeling Sponsored: Telecommunications and Network Analytics Sponsored Session Chair: Eli Olinick, Southern Methodist University, Dallas, TX, 75275- 0123, United States 1 - Big Data Processing Platform for Healthcare Applications Albena Mihovska, Aarhus University BSS, Birk Centerpark 15, Herning, 7400, Denmark This work proposes a novel conceptual approach to IoT/cloud convergence and its implementation in healthcare application platforms. The proposed platform allows for the collection, exchange and processing of healthcare related data without distorting the quality of the collected data and without compromising the personal aspects of the processed data; based on an elastic from point of view of security and performance wireless network and a distributed processing platform and their seamless integration. To this end the work addresses the processing of multiple data streams, the design of the interfaces between the sensing and cloud environments; and the secure communications between the IoT devices and big data applications. 2 - Impact of Telecommunications on Household Finances Stanko Dimitrov, Associate Professor, University of Waterloo, 200 University Avenue West, Waterloo, ON, N2L 3G1, Canada, Brian Paul Cozzarin Telecommunications have positive impacts such as reduced cost of data delivery. However, there may be negative, financial, impacts. Using a Canada wide survey data, we find that using online banking causes an increase in personal debt-to- income ratio of 1.42 dollars more for every dollar earned on average, relative to not banking online. Our finding is supported and contributes to work in telecommunications in retail. The findings hopefully inform consumers’ actions when deciding to engage in online banking, as well as banks when assessing the risk of its customers. n MD61 West Bldg 102C

3 - Determining Queue Assignment and Threads in a Multi-Queue Multi-Threaded System Adam Colley, Southern Methodist University, Dallas, TX, United States, Eli Olinick We present a methodology for performance tuning an online business-to-business integration (B2BI) service that manages a system of G/G/1 queues running on a multicore computer processing unit (CPU), or network of CPUs. Tuning consists of optimizing the assignment of processes to priority queues, the assignment of threads to queues, and the relative queue weights. We demonstrate the methodology, and analyze its performance, on a real-world application of the IBM B2B Integrator.

n MD62 West Bldg 103A Interfaces Between Optimization and

Machine Learning Sponsored: Data Mining

Sponsored Session Chair: Velibor Misic 1 - Discovering Optimal Policies: A Pattern Recognition Approach to Model Analysis Fernanda Bravo, UCLA Anderson School of Management, 110 Westwood Plaza, B4.11, Los Angeles, CA, 90024, United States, Yaron Shaposhnik We study the application of machine learning algorithms to derive insights and identify structural properties of mathematical models. As a proof of concept, we apply the proposed framework to core optimization problems, such as inventory replenishment, queueing admission control, revenue management and multi- armed bandit problems. We formulate classification and regression problems and show how numerically obtained optimal solutions can be used to identify structural properties, derive compact representations and interpret complex policies. 2 - Optimization of Tree Ensembles Tree ensemble models such as random forests and boosted trees are often used to predict the effect of different decisions. While such models are widely used for predictions, little is known about how to use them for effective decisions. In this talk, we consider the problem of finding a decision that is optimal with respect to a tree ensemble model. We formulate this problem as a mixed-integer optimization model and present theoretical results on its structure. We test our method on real data sets, including two case studies in drug design and customized pricing, and show that it can efficiently solve large-scale instances to near or full optimality, outperforming heuristic solutions. n MD63 West Bldg 103B Representation Learning for Unstructured Data Sponsored: Data Mining Sponsored Session Chair: Mustafa Gokce Baydogan, PhD, Bogazici University, Bogazici Universitesi, Endustri Muh. Bolumu, Istanbul, 34342, Turkey 1 - A Dissimilarity Regularization Approach for Distinct Feature Learning with Autoassociative Neural Networks Phillip Howard, Intel Corporation, 5000 W. Chandler Blvd, Chandler, AZ, 85226, United States, Daniel Apley, George Runger Autoassociative neural networks (ANNs) are useful for performing dimensionality reduction and have been proposed as a nonlinear extension of principal component analysis. While ANNs can learn complex nonlinear representations of high-dimensional data, they also have a tendency to extract repetitive features which are difficult to interpret. In this work, we present a new dissimilarity regularization method for encouraging the features learned by ANNs to be distinct. We demonstrate the effectiveness of our method for learning distinct Velibor Misic, UCLA Anderson School of Management, 110 Westwood Plaza, Suite B406, Los Angeles, CA, 90095, United States

236