2016 INFORMS Annual Meeting Program

WD01

INFORMS Nashville – 2016

Wednesday, 2:45PM - 4:15PM

3 - Control System For Electronic Triage In The Emergency Department: Integrating The User Into Development Loop Diego A. Martinez, Johns Hopkins School of Medicine, Baltimore, MD, 21201, United States, dmart101@jhmi.edu, Scott R Levin The potential for machine learning systems to improve via exchange of information with knowledgeable users has yet to be explored in much detail. In a pilot study in an emergency department of a large hospital, nurses were presented with triage level predictions, and they were able to provide feedback through a real-time communication system. The types of some of this feedback seem promising for assimilation of clinical gestalt by machine learning systems. The results show that to benefit from clinical gestalt; machine learning systems must be able to absorb information in a graceful manner and provide clear explanations of their predictions. WD03 101C-MCC Big Data IV Contributed Session Chair: Wenbo Sun, University of Michigan, 2013 Medford Rd Apt 161, Ann Arbor, MI, 48104, United States, sunwbgt@umich.edu 1 - A Revaluation Of The Relationship Between Environmental Management And Financial Performance – A Multilevel Longitudinal Analysis Zuoming Liu, Lynchburg College, 1501 Lakeside Drive, School of Business & Economics, Lynchburg, VA, 24501, United States, lzuoming@gmail.com This study employs a multilevel cross-lagged model to investigate the causal relationships between a corporation’s environmental performance and financial returns by using a 4-year dataset of the largest US500 companies. The goal is to identify variation of relationship due to different features at various levels. By conducting multilevel analyses, the relationship between environmental performance and financial returns is demystified into three levels, intra-firm dynamic variations over time, inter-firm variations, as well as variations across industries. 2 - A Class Of First Order Methods That Do Not Rely On Any Norm This work generalizes the notion of smoothness, strong convexity, and Lipschitz continuity of a convex function by introducing a reference function, and uses the reference function to derive convergence rates for generalized first-order methods for convex optimization. The approach yields clear intuition behind convex optimization with composite functions as a corollary. We also developed a first- order interior-point method using a weak definition of self-concordance. 3 - A Method For Developing Confidence Bands For Multiple Dimensional Functional Responses Wenbo Sun, University of Michigan, 2013 Medford Rd Apt 161, Ann Arbor, MI, 48104, United States, sunwbgt@umich.edu, Jionghua Jin The standard method for specifying target responses’ variabilities involves developing a confidence band for a set of empirical mechanical responses. These responses are multiple-dimension signals obtained from identical trials of different subjects. The existing methods commonly normalize responses with respect to subject characteristics, point-wisely generate confidence bands ignoring times and direction’s correlation. A new method was developed in the structure of mixture models based on basis-representation and Gaussian process and provided an approach for outlier detection. It is applied to the kinematic response data collected in Children’s Hospital of Philadelphia. Haihao Lu, PhD Student, MIT, 60 Wadsworth St, Apt 8B, Cambridge, MA, 02142, United States, haihao@mit.edu, Yurii Nesterov, Robert Michael Freund

WD01 101A-MCC Data Mining Application in Business Sponsored: Data Mining Sponsored Session Chair: Parvaneh Jahani, University of Louisville, 781 Theodore Burnett Court, Apt 2, Louisville, KY, 40217, United States, parvaneh.jahani@louisville.edu 1 - Machine Learning And Cognitive Pricing Zhengliang Xue, IBM Research Center, 1101 Kitchawan Road, Route 134, Yorktown Heights, NY, 10598, United States, zxue@us.ibm.com, Markus Ettl We study a method to price personalized configuration of software and services. The seller has to deal with a customized configuration without any similar records in history. A data-driven approach is applied to estimate the purchase probability for any unique configuration based on historical trading data. In addition, client relationship and firmographic information need to be incorporated to the pricing decision. We establish a utility model to evaluate the configuration and recognize the impact of relationship. The business impact of optimal pricing is justified by the actual data. 2 - Investigating Sparse Demand Models To Support The Assortment Planning Decision Matthew Lanham, Purdue University, West Lafayette, IN, 47905, United States, malanham@gmail.com, Ralph D Badinelli We present research examining the performance of substitution-based multi- classification models currently being researched and employed in practice by major retailers, versus more naïve binary classification models to understand purchase propensity. We discuss how these models would yield different assortments for sparse demand products. WD02 101B-MCC Decision Analysis in Health Care Data Mining Sponsored: Data Mining Sponsored Session Chair: Diego Martinez, Johns Hopkins University, 1, Baltimore, MD, 212, United States, dmart101@jhmi.edu 1 - Multiscale Decision Making Based on Variational Evidential Reasoning For Medical Record Label Recommendation Haiyan Yu, Lecturer, Chongqing University of Posts and Telecommunications, 2 Chongwen Road, Nan’an District, Chongqing, P.R.C, 400065, China, yhy188@tju.edu.cn Haiyan Yu, Lecturer, University of Electronic Science and Technology of China, Chengdu, China, yhy188@tju.edu.cn, Man Xu, Jiang Shen Due to the characteristics of fragmentation and uneven distribution of the clinical data, the issues of reducing its misdiagnosis and improving the system discrimination ability arises in evidential reasoning. To solve these issues, the model of belief propagation was constructed based on the hierarchical Dirichlet process, achieving the confidence fusion of fragmented evidence. Through parameter learning, the recommendation of indentifying the lables of medical instances were achieved with the control trajectory of the reasoning model. Finally, simulation study verified the effectiveness and resilience of the reasoning models, improving the efficiency and quality of medical services. 2 - Using Matrix-based Multi-criteria Decision Method For Assessing Risk Of Harm Of Alert-overridden Intravenous Infusions Wan-Ting Su, PhD Student, Purdue Univeristy, West Lafayette, IN, United States, su33@purdue.edu, Poching DeLaurentis, Mark Lehto Hospital medication safety teams regularly review and analyze infusion drug alert reports to evaluate infusion performance. Previously an Intravenous (IV) medication harm index was developed to help clinicians assess the potential patient harm of each alert-overridden infusion. We aim to apply a matrix-based multi-criteria decision method to improve the existing harm index. The improved index can help medication safety teams better identify the medication and care unit combinations of high risk and further prioritize the targets for improvement on nursing practice, workflows and drug limit settings.

456