Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

MC66

4 - A Novel, Scalable Machine Learning Task Capable of Superhuman Artificial Intelligence Evan Barlow, Weber State University, Goddard School of Business & Economics, 1337 Edvalson St, Ogden, UT, 84408, United States Machines often outperform humans in executing repetitive physical tasks. Leveraging machine learning, computers can now: (i) mimic human performance in repetitive mental tasks, or (ii) exceed human performance in complicated predictive tasks. Prediction is most fitting, however, when inputs and outputs are out of decision makers’ control. Large-scale prescriptive analytics using optimization is very difficult (if possible at all) for most current machine learning tasks. The machine learning task presented here promises a scalable approach to superhuman decision making. The approach has successfully been applied to prescribe optimal decisions with several large-scale sandbox datasets. Joint Session AI/Practice Curated: Healthcare Analytics: Machine Learning Approaches for Health Data Sponsored: Artificial Intelligence Sponsored Session Chair: Hongyi Zhu 1 - Automatic Diagnosis of Alzheimer’s Disease Using Deep Neural Networks Maryam Zokaeinikoo, Graduate Research Assistant, Pennsylvania State University, University Park, State College, PA, 16802, United States, Prasenjit Mitra We propose different neural network models based on Long Short-Term Memory (LSTM) to detect the onset of Alzheimer’s early in the course of the disease using textual data from both healthy subjects and patients. These models include LSTM networks, bidirectional LSTM (BLSTM), bidirectional LSTM with attention layer (Attention-BLSTM), and bidirectional LSTM with conditional random fields layer (CRF-BLSTM). Although the LSTM often requires large training datasets, our CRF-BLSTM algorithm demonstrates that even with limited training data it performs well in detecting the Alzheimer’s disease. The results are validated using two methods of cross validation. 2 - Computational Algorithms for Tracking Near Falls with Multiple Wearable Sensors Alla Kammerdiner, New Mexico State University, Las Cruces, NM, United States, Razan Ayasra A loss of balance that constitute near falls can be tracked with multiple body- worn accelerometers. We consider some new formulations for estimation and optimization problems related to tracking of near falls. We also present and analyze computational algorithms, which are used for space partitioning in statistical estimation and for solving the combinatorial optimization problems. 3 - Seizure Detection Using a Hidden Markov Model Framework Mahboubeh Madadi, Louisiana Tech University, College of Engineering and Science, P.O. Box 10348, Ruston, LA, 71272, United States, Giovanni Petris, Leonidas Iasemidis In this study, a hidden Markov models (HMM) is developed to automatically detect different brain states in epilepsy patients. The proposed HMM aims at characterizing the dynamics of intracranial electroencephalographic (iEEG) signals using important features such as generalized partial directed coherence (GPDC). The iEEG signals from epilepsy patients, who underwent long-term monitoring of the brain electrical activity for subsequent surgical removal of their epileptogenic focus, were analyzed to classify a patient’s state at a given point in time into one of four states: interictal (seizure-free / between seizures), pre-ictal (pre-seizure), ictal (seizure), postictal (post-seizure). 4 - A Deep Learning Approach of Microarray Data Analysis in Cancer Prognosis Sharmin Nahar Mithy, Doctoral Candidate, University of South Florida, Tampa, FL, 33612, United States, Grisselle Centeno Research has shown that a major portion of cancers and related deaths could be prevented by applying existing knowledge about cancer treatment. In this research we present a deep learning approach for cancer gene identification. SDAE based clustering approach is applied here for feature extraction from Gene Expression Profiling and will be compared with some other methods. The performance of the extracted information will be evaluated to verify the usefulness of the new features. The extracted feature will further be utilized to predict the radio sensitivity for 48 cell line used as input in cancer treatment and patient prognosis. We expect to validate the data from 20 patients with rectal cancer. n MC66 West Bldg 105A

n MC67 West Bldg 105B Platforms and Peer-to-Peer Markets Sponsored: Information Systems Sponsored Session Chair: Zaiyan Wei, Purdue University, Purdue University, West Lafayette, IN, 47907, United States 1 - Is Home Sharing Making Housing Less Affordable? Evidence from a Natural Experiment on Airbnb Wei Chen, Assistant Professor, University of Arizona, 1130 East Helen Street, Tucson, AZ, 85721, United States, Zaiyan Wei, Karen Xie We study the impact of online home sharing on affordable housing. We leverage a “natural experimentöùa platform regulation that caps the number of properties a host can manage in some marketsùto estimate the impact. We find that the restriction was associated with a 3.5% decrease in local rental prices and a 1.9% decrease in housing value. The decrease can be attributed to the removal of absentee landlords’ properties from Airbnb back to local markets. The price-to- rent ratio, however, increased by about 1.6%, which suggests that online home sharing is mainly a substitute for local rental markets. These findings speak to the question whether home sharing makes housing less affordable. 2 - Does Political Polarization Decrease Market Efficiency: An Investigation in the Context of Online Lending Hongchang Wang, Georgia Institute of Technology, 800 West Peachtree NW, Atlanta, GA, 30308, United States, Eric Overby We study whether political polarization inhibits market efficiency by examining whether investors in online lending markets are less likely to lend to borrowers whose political ideology (i.e., liberal or conservative) is likely to be different from their own. We apply both a gravity model and a difference-in-differences model to find that borrowers in liberal states are more preferred than borrowers in conservative states (on average 5% more bids). In addition, borrowers are more likely to attract investors from politically similar states than investors from politically dissimilar states. We find that political distance deters online lending, with an impact of 1.8% for one standard deviation. 3 - The Roles of Informed and Uninformed Backers on Crowdfunding Platforms Aravinda Garimella, University of Washington, Foster School of Business, Mackenzie 350, Seattle, WA, 98105, United States Backers of crowdfunding projects are heterogeneous in market knowledge, sophistication and overall informedness. We examine the roles of informed and uninformed backers on reward-based crowdfunding platforms. Using transaction- level data from a leading Chinese crowdfunding platform, we classify backers into informed and uninformed backers based on their contribution patterns. We study how uninformed backers affect fundraising dynamics and the ultimate performance of crowdfunding projects. Drawing from theoretical models of investor sentiment, we offer findings with important implications for crowdfunding platforms and entrepreneurs. 4 - Is Non-persistent Social Status a More Useful Incentive Mechanism? Evidence from Yelp Elite Squad Mingyue Zhang, Beijing Foreign Studies University, Beijing, China, Xuan Wei Content sharing platforms such as product review websites largely depend on users’ voluntary contributions. To motivate the contributions, many platforms established reputation-based incentive mechanisms. Yet most of the existing research has focused on reputations that are everlasting. In this research, we study the effect of non-persistent social status on user behavior using data from Yelp Elite Squad which is a yearly program. We design fixed effect model as well as matching method to empirically study the effect of non-persistent social status on user’s behavior in both short term and long term. Our study has significant implications for business models that rely on user contributions. 5 - Do You Donate Out of Altruism or Self-interest? A Case of Conditional Social Gift Exchange in Fundraising Cenying Yang, University of Texas at Austin, Austin, TX, 78703, United States, Shun-Yang Lee, Andrew B. Whinston In this paper we explore the motivation, altruism or self-interest, behind donation in the context of social gift. We tease out one’s altruistic motivation from the self- interest motivation by introducing different levels of recommended donation amount through a randomized experiment. The results show that people behave selfishly when a social gift is incorporated in a more altruistic context, but altruistically when it is incorporated in a more self-interest context

208