Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SA66

3 - Survival Rate Prediction in Cardiac Patients with Heart Transplant or Assisted Devices Maryam Soltanpour Gharibdousti, Binghamton University, 1120 Murray Hill Road, Vestal, NY, 13850, United States The survival rate prediction for the organ transplant surgery patients can help to classify patients risk levels and potential post-surgical complications. The research used the data for cardiac patients with either medical assist devices such as Impella and Left Ventricular Assist Devices (LAVD) or heart transplant patients. The significant factors such as demographic information, baseline patient characteristics, baseline hemodynamics, laboratory values, and in-hospital complications can predict the survival rate after the transplant surgery. The data from one of the Organ Procurement Organizations (OPO) in New York state is analyzed using several machine learning algorithms. n SA64 West Bldg 104A Joint Session DM/AI/pratice Curated: Urban Big Data Analytics and Mining Sponsored: Data Mining Sponsored Session Chair: Xun Zhou, University of Iowa, Iowa City, Iowa 1 - Predicting Urban Dispersal Events: A Two-stage Framework Through Survival Analysis Amin Vahedian Khezerlou, University of Iowa, IA, United States In this work, we focus on predicting unexpected dispersal of people in urban setting, based on taxi pick-up records. Unlike regular taxi pick-up patterns, which are highly regular and predictable, the irregular dispersal events do not follow an obvious pattern and are challenging to predict. Such predictions can be used to better plan public safety and traffic management, as well as business profit. We propose a supervised learning framework which takes advantage of survival analysis to infer the event time, and also predict the event volume. We use a public dataset of taxi records for evaluations. We outperform baseline methods by a significant margin. 2 - Multi-Frequency Convolutional LSTM for Crime Prediction Maryam Rahmani Moghaddam, Univeristy of Iowa, Iowa City, IA, United States The crime prediction problem aims at predicting the location and time of the future crimes. This problem is challenging due to the spatial and temporal sparsity of the crimes. In this work, we propose a Multi-Frequency Convolutional LSTM model, which is an ensemble of multiple convolutional lstm models trained by subsets of historical data with varying sampling intervals to capture the multi- frequency patterns of crimes. The combined output of these models using a convolutional neural network is further combined with the historical map through a spatial regression. We test our model on 5 years (Feb 2015-Feb 2017) of burglaries (23797 reports) in Portland, Oregon provided by the Portland Police Bureau and evaluate the PAI and PEI. We compare the proposed model with baseline methods static-map and Self-Exciting Point Process. 3 - A Holistic Solution for Connected Smart City with Good Traffic Management is very important aspect towards smart city. As a solution towards smart city, this work concentrates on multiple aspects like 1) IOT based systems adapting Deep Learning and Machine Learning Techniques to accurately identify vehicles that violate traffic rules 2) Estimate the probability of a parking space being available in a locality 45 mins in advance and update with time 3) Estimate the impact of traffic realignment on the environment and iterate accordingly. Overall the solution aims towards a connected city and bring a balance between the environmental health and comfort towards traffic movement and rules in a holistic view. Environmental Health, Traffic and Energy Sheela Siddappa, Bosch, Bengaluru, India

method utilizes the heart rate, acceleration and speed data automatically collected by the experimenters when they are walking in subway transfer stations, fits these data to Physical Activity Intensity and uses it as the index of travel energy cost. Subsequently, the accuracy, theoretical and practical prospect of this method are verified by the transfer passenger data of Beijing Subway Line 1 and Line 2 in May, 2017. The results show that the service resilience calculation method can accurately perceive the change of system service efficiency and its recovery ability according to different travel demands of the passengers. At the same time, the method uses automatic data collection to analyze, improves its accuracy and analysis adaptability compared with the traditional analysis methods. . 2 - Does User Contribution Enhance Welfare? The Effectiveness of User-crowdsourced Content in Relieving Urban Traffic Congestion Tae Hun Kim, Baylor University, Waco, TX, 76798, United States, Chenhui Guo, Anjana Susarla, Vallabh Sambamurthy In a mobile virtual community, Waze app users generates primary information (via alerts), follow-up feedback (via comments), and collective confirmation (via thumbs-up). We examine whether and how the user-crowdsourced content generates welfare by relieving traffic jams in New York City. Spatial panel data models are applied to estimating welfare value with large-scale data on behaviors and locations of users. Based on the results, user contribution reduces traffic congestion duration. A welfare analysis estimates that user contribution saves 2.28% of annual congestion cost per driver in the city. The welfare value shows that user contribution is effective in saving social and economic costs. n SA66 West Bldg 105A Artificial Intelligence in Big Data Sponsored: Artificial Intelligence Sponsored Session Chair: Jiaheng Xie, University of Arizona, Tucson, AZ, 85721, United States 1 - Visual Social Media Analytics: Impact of Medical Knowledge on User Engagement Xiao Liu, University of Utah, UT, United States Video sharing social media sites, such as YouTube, that host videos providing information on the pathogenesis, diagnosis, treatments, and prevention of various conditions can be an effective way to understand medical knowledge and in managing chronic conditions through patient self-care. However, due to the heterogeneity of the content quality and content helpfulness on visual social media, healthcare providers and government agencies have expressed concerns about the quality and reliability of such information. There have been relatively few studies that have identified interventions to increase the ease with which patients can find helpful health information. We propose an interdisciplinary lens that synthesizes deep learning methods with themes emphasized in Information Systems (IS) research and research on healthcare informatics. Using a bidirectional long short-term memory (BLSTM) method, we extract medical terminology from videos. We annotate videos using inputs from domain experts and build a logistic regression based classifier to categorize videos based on whether they encode a high degree of medical knowledge or not. We identify distinct types of user engagement with videos on YouTube using a principal components analysis (PCA) approach: user dissonance, popularity based engagement, and relevance based engagement. We find that medical knowledge encoded in videos matters to patient engagement; however, popularity-based indicators of engagement indicate that videos that score high on medical knowledge encoded in videos, are actually less popular than those that are not. We conduct robustness checks using a convolutional neural network (CNN) to detect the presence of medical objects in a video. We find that medical terminology embedded in textual data is more salient to an assessment of medical knowledge encoded in a video, rather than image analytics. Our results suggest that healthcare practitioners and policymakers need a nuanced understanding of how users engage with medical knowledge in video format, which has implications for the role of videos and visual social media in bridging the health literacy gap and in enabling self-care of chronic conditions. 2 - Are More Diverse Crowds Smarter? Jiayu Yao, Georgia Institute of Technology, Atlanta, GA, United States, Qiang Gao, Mingfeng Lin Does the diversity increase the wisdom of crowds? We examine the value of diversity in crowds within a market setting, specifically, the online financial market, utilizing a natural experiment on Prosper.com.

n SA65 West Bldg 104B Big Data Science Sponsored: Data Mining Sponsored Session

Chair: Tae Hun Kim, Michigan State University, Eli Broad College of Business, 632 Bogue Street, East Lansing, MI, 48824, United States 1 - Analysis of the Resilience of Public Transportation System Based on Energy Cost of Passengers Xiongfei Lai, Dr., Tongji University, No. 4800, Cao’an Road, Shanghai, 201804, China, Jing Teng, Lu Ling Based on field experiments, this paper proposes a calculation method for public transportation service resilience based on the energy cost of passengers. This

25