2016 INFORMS Annual Meeting Program

SD02

INFORMS Nashville – 2016

Keynote Davidson Ballroom C-MCC Analytic + IT + Deployment = Real World Impact Keynote Session Robin Lougee, IBM TJ Watson Research Center, Yorktown Heights, NY 10598, rlougee@us.ibm.com 1 - Analytic + IT + Deployment = Real World Impact Ramayya Krishnan, Carnegie Mellon University, Pittsburgh, PA, United States, rk2x@cmu.edu The Heinz College is home to two highly ranked graduate schools: 1) Information Systems and Management and 2) Public Policy and Management, a deliberate structure which exists only at Carnegie Mellon University (CMU). Founded by noted Management Scientist W. W. Cooper to educate “men and women capable of intelligent action”, the unique structure of the college gives its educational programs a holistic focus on societal problem solving. This focus translates into teaching cutting-edge information technologies and analytic methods and providing students with multiple opportunities to apply them to solve real world problems that matter. This focus also means an emphasis on structuring unstructured problems and an education in the skills required to be effective at that structuring and at decision making, and engendering change through deployment. In this keynote, I will provide an overview of our award winning analytics program and describe how we combine industry-funded research centers and their partner ecosystems to provide students with multiple opportunities to learn an array of analytic skills and problem-solving expertise in order to be effective in the real world. SC94 5th Avenue Lobby-MCC Wiley/Provalis Research Technology Tutorial: Wiley/Provalis 1 - Provalis Research will Showcase its Integrated Collection of Text Analytics Software Normand Peladeau, Provalis Research, Montreal, QC, Canada. peladeau@provalisresearch.com 2 - Wiley: Interested in Publishing for the Wiley Series in Operations Research and Management Science? James Cochran, University of Alabama, Tuscaloosa, AL, jcochran@cba.ua.edu The Wiley Series in Operations Research and Management Science is a broad collection of books that meet the varied needs of researchers, practitioners, policy makers, and students who use or need to improve their use of analytics. The workshop will include presentations on the following: • The Mission of the Series for the Betterment of the Community • The Proposal Process: Maximizing your Success • What’s Next: The Writing, Review, and Publishing Process • Q&A In addition to the presentation, you will be able to meet with Founding Series Editor James J. Cochran and Wiley Editor Susanne Steitz-Filler to discuss any further questions or potential book ideas you may have. SD01 101A-MCC Interpretable Machine Learning in Social Science Sponsored: Data Mining Sponsored Session Chair: Tong Wang, MIT, Cambridge, MA, United States, tongwang@mit.edu 1 - Interpretable Decision Sets: A Joint Framework For Description And Prediction Himabindu Lakkaraju, Stanford University, himalv@cs.stanford.edu One of the most important obstacles to deploying predictive models is the fact that humans do not understand and trust them. In this talk, I will present interpretable decision sets, a framework for building predictive models that are highly accurate, yet also highly interpretable. We formalize decision set learning via an objective function that simultaneously optimizes for accuracy, conciseness, and meaningfulness of the rules. We prove that our objective is a non-monotone submodular function, and efficiently optimize it with a 2/5 approximation guarantee. Our experiments demonstrate that interpretable decision sets help humans understand their data better than other interpretable models. Abstract Available Online. Sunday, 4:30PM - 6:00PM

2 - Exploring Complex Systems Using Semi-parametric Graphical Models

Mladen Kolar, University of Chicago, mkolar@chicagobooth.edu Extracting knowledge and providing insights into the complex mechanisms underlying noisy high-dimensional data sets is of utmost importance in many scientific domains. Networks are an example of simple, yet powerful tools for capturing relationships among entities over time. For example, in social media, networks represent connections between different individuals and the type of interaction that two individuals have. Unfortunately the relationships between entities are not always observable and need to be inferred from nodal measurements. I will present a line of work that deals with the estimation and inference in high-dimensional semi-parametric elliptical copula models. 3 - Causal Rule Set For Subgroup Identification Tong Wang, University of Iowa, tongwang@mit.edu We propose an interpretable classifier for causal analysis, Causal Rule Set (CRS), that discriminates between two types of subgroups, one that benefits from the treatment (Effective Class), and one does not (Ineffective class). CRS uses a set of interpretable rules to present and characterize an Effective class. We present a Bayesian framework with priors that favor simple models, and a Bayesian logistic regression that models the relation between outcomes and a set of observed (attributes) and inferred objects (subgroup membership). The simulation studies and experiments on real data sets show that distributing treatment according to a CRS model enhances the average treatment effect. SD02 101B-MCC Data Analytics and Modeling for Medical Prognosis and Decision Making Sponsored: Data Mining Sponsored Session Chair: Shouyi Wang, University of Texas-Arlington, Arlington, TX, United States, shouyiw@uta.edu 1 - Disgnosis Of Posttraumatic Stress Disorder Using Functional Near Infrared Spectroscopy (fNIRS) signals And Data Mining Techniques Rahil Hosseini, University of Texas at Arlington, rahilsadat.hosseini@mavs.uta.edu In this paper we extract various feature groups from FNIRS records that are from the experiment about digits memorizing and recalling; it includes three phases in each trial; encode, maintain and recall; we show the discovered patterns between two classes for some selected features. Specifically the results show that the last phase which is when the subject tries to recall the digits, is the most significant part and with extracted features from Statistics, Autocorrelation and SVDNorm; it is enough for highly accurate classification.We discuss a new proposed feature derived from SVD (Singular Value Decomposition) of raw data in channels. It demonstrated promising results in classification. Third contribution is comparison of feature selection methods to reduce the dimension of the feature matrix. We compare the performance through number of selection and sensitivity and specificity and their average. The proposed method includes Mutual Information (MI) guided sparse models that outperform the existing features selection techniques. The existing models are ‘’minimum Redundancy and Maximum Relevancy’’ (mRMR), and ‘’Sparse Group Lasso’’ (SGL). We propose ‘’Mutual Information and Least Absolute Selection and Shrinkage Operator’’ (MILASSO) ‘’Mutual Information Sparse Group Lasso’’ (MISGL). All these techniques are applied to classify PTSD veterans and healthy controls using ‘’Support Vector Machines’’ (SVM). Last contribution is finding the Region of Interest (ROI), we conclude that two specific areas on brain are the most significant ones which are directly related to memorizing 2 - Pattern Classication And Analysis Of Memory Processing In Depression Using Eeg Signals Kin Ming Puk, University of Texas at Arlington, bookbook0089@gmail.com An automatic, EEG-based approach of diagnosing depression with regard to its effect on memory is presented. EEG signals are extracted from 15 depressed subjects and 12 normal subjects during experimental tasks of reorder and rehearsal. After pre-processing noisy EEG signals, seven groups of mathematical features are extracted and classification with SVM is conducted under a five-fold cross-validation, with accuracy of up to 70% - 100%. The contribution of this paper lies in illustrating the usefulness of the classification framework in facilitating the analysis and visualization of the difference of EEG signals between depressed and control subjects in memory processing.

95