Informs Annual Meeting 2017

TD71

INFORMS Houston – 2017

TD69

3 - Classification of Intracardiac Electrogram During Atrial Fibrillation using Hidden Markov Models and Gaussian Mixture Models Lianning Zhu, Texas Tech University, 2500 Broadway, Lubbock, TX, 79409, United States, lianning.zhu@ttu.edu, Dongping Du This talk presents a new classification algorithm to identify electrical triggers in Atrial Fibrillation using hidden Markov models(HMM) and Gaussian mixture models(GMM). Activations of multi-lead Intracardiac Electrogram(IEGM) are detected based on the variance equality test. Wavelet transform is used to denoise and extract features to train HMMs. HMMs are developed for each lead of signal and used to calculate the log-likelihood of each activation in the IEGM. Computer simulation is performed to generate modeled IEGM for model testing and validation. The accuracy and sensitivity of this diagnosis system are relatively high. 4 - Robust Principal Component Analysis in Cloud-based Run-time Environment Kyungduck Cha, SAS.Institute Inc, Sas Campus Drive, Cary, NC, 27513, United States, Kyungduck.Cha@sas.com, Seunghyun Kong, Zohreh Asqharzadeh, Arin Chaudhuri Robust Principal Component Analysis (RPCA) is an enhanced statistical procedure to perform the principal component analysis in the presence of noisy data. Principal component analysis is widely used in data mining and machine learning. RPCA aims to recover a low rank matrix from grossly corrupted observations. RPCA has been applied in many areas, including image processing, latent semantic indexing, ranking and recommendations, and anomaly detection. The RPCA techniques have been implemented in SAS Viya platform, the cloud-based run-time environment for data management and analytics. We present the advantage of the distributed computing and how to use clients other than SAS, such as Python 5 - A Data Framework of Financial Deterioration Detection for Chinese Firms Jun Huang, assistant professor, San Angelo State University, San Angelo, TX, United States, jun.huang@dusty.tamiu.edu, Haibo Wang Firm level credit risk prediction, such as financial deterioration detection or financial distress prediction, is used to mitigate the risk of non-performing loans. The objectives of this study include: (1) Identify most efficient model in predicting the financial deterioration of Chinese firms based on different statistical and machine learning algorithms; (2) Use the model(s) to establish a warning mechanism to detect financial deterioration in an early stage; (3) Identify effective financial indicators for the prediction model by using a feature selection technique; (4) Test whether discretization of financial ratios improve the prediction performance of this classification problem. 371F 2:00 - 2:45 SAS/ 2:45 - 3:30 JMP, a Division of SAS Invited: Vendor Tutorial Invited Session 1 - Building and Solving Optimization Models with SAS Edward P. Hughes, SAS.Institute, Inc., Sas Campus Drive, Cary, NC, 27513, United States, ed.hughes@sas.com, Rob Pratt SAS provides a broad and deep array of data and analytic capabilities, including data integration, statistics, data and text mining, econometrics and forecasting, and operations research. The SAS optimization, simulation, and scheduling features coordinate easily and fully with other SAS strengths in data handling, analytics, and reporting. OPTMODEL from SAS provides a powerful and intuitive algebraic optimization modeling language and unified support for building and solving LP, MILP, QP, NLP, CLP, and network-oriented models. And because the OPTMODEL optimization modeling language is contained within the OPTMODEL procedure, a SAS software module, it integrates seamlessly with the entire family of SAS functions, procedures, and macros. We’ll demonstrate how you can use OPTMODEL to solve both basic and advanced problems, highlighting its newer capabilities and its support for both standard and customized solution strategies. 2 - JMP Pro: Top Features That Make You Say “Wow”! Mia L. Stephens, SAS.Institute Inc, P.O. Box 290, York Harbor, ME, 03911, United States, mia.stephens@jmp.com JMP is data visualization, analysis and modeling software from SAS for Mac or Windows. Come see the top 10 features in JMP Pro 13 that JMP users are raving about, including Text Explorer (for preparing and analyzing unstructured text data), the JMP Query Builder (for generating queries and writing SQL code directly in JMP), the Formula Depot (for comparing predictive models, scoring new data and generating C, Python, SAS, SQL and JavaScript code), dashboards, interactive web reports, and the latest visualization techniques in the drag-and- drop Graph Builder. TD71

371D FSS Student Paper Competition Sponsored: Financial Services Sponsored Session

Chair: Tim Siu-Tang Leung, University of Washington, Dept of Applied Mathematics, University of WA, Lewis Hall #202, Seattle, WA, 98195- 3925, United States, timleung@uw.edu 1 - Clearinghouse Default Waterfalls: Risk-sharing, Incentives, and Systemic Risk Allen Cheng, Columbia University, New York, NY, 10027, United States, wc2232@columbia.edu We model the clearinghouse waterfall, explain empirical capitalization differences, and provide regulatory insights. 2 - Portfolio Construction by Mitigating Error Amplification: The Bounded-Noise Portfolio Long Zhao, UT McCombs Business School, 2110 Speedway Stop B6500, CBA 5.334 B, Austin, TX, 78712-1277, United States, longzhao@utexas.edu By dissecting covariance estimation error, a new portfolio achieves great performance in both simulation and real data. 3 - Default Risk Premia and a Non-Linear Asset Pricing Model Yi Marco Zhang, University of Chicago, Booth School, Chicago, IL, United States, marco.y.zhang@chicagobooth.edu The value of an equity investment can be framed as an embedded call option on a firm’s assets, and losses given default create discontinuities in the value of these options. In this paper, we introduce a non-linear equity pricing model that includes these aspects of potential default consequences. 4 - A Dynamic Network Model of Interbank Lending Xu Sun, Columbia University, 363 W 123rd St Apt 4R, New York, NY, 10027, United States, xs2235@columbia.edu We develop a dynamic model of interbank borrowing and lending activities in which banks are organized into clusters, and adjust their monetary reserve levels to prescribed target. We model the transactional dynamics through a set of interacting measure-valued processes and establish the weak limit of the interacting measure-valued processes as the number of banks grows large. 371E Data Mining Contributed Session Chair: Kyungduck Cha, SAS Institute Inc, Cary, NC, United States, Kyungduck.Cha@sas.com 1 - One-class Relaxed Support Vector Machines Petros Xanthopoulos, Stetson University, 421 Woodland Blvd, DeLand, FL, 32723, United States, pxanthopoulos@stetson.edu, Orestis P. Panagopoulos, Athanasios A. Panagopoulos, Talayeh Razzaghi, Onur Seref In this work, a new novelty detection method called One-Class Relaxed Support Vector Machines (ORSVMs) is proposed for datasets with outliers. ORSVMs are formulated using a quadratic loss function and are solved with sequential minimal optimization. Their performance is measured on several publicly available datasets and are compared to other state-of-the-art one-class classification methods. 2 - Lexicographic Multiobjective Programming (LMOP) Models for Feature Selection in LAD Cui Guo, Associate Professor, Shantou University, 243 Da Xue Road,, Shantou, Guangdong Province, 515063, China, cguo@stu.edu.cn, Kedong Yan, Peihua Zhao The bianrization step of Logical Analysis of Data(LAD) replaces each numerical attributes with a set of binary variables. This introduces large amount of binary features, which increases the complexity of subsequent pattern generation procedure. A feature selection method based on set covering was proposed to minimize the number of binary variables. In practice, however, it’s also expected to reduce the number of real attributes. In this presention, we’ll introduce LMOP models to select the optimal sets of variables w.r.t. to different lexicographic preferences. Theories and their proofs will be given. And the utilities of the propsed models will be demonstrated on various benchmark datasets. TD70

401