Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TB43

2 - Insights Revealed by the Value of Information in a Multiple- objective Decision: Brown Trout in the Grand Canyon Michael C. Runge, USGS Patuxent Wildlife Research Center, 12100 Beech Forest Rd, Laurel, MD, 20708, United States, Charles B. Yackulic, Lucas S. Bair, Theodore A. Kennedy, Richard A. Valdez Brown trout (Salmo trutta) have expanded their range in the Colorado River ecosystem, possibly threatening a delicate balance among multiple objectives, including recovery of the endangered humpback chub (Gila cypha) and ecosystem restoration. We examined possible causal mechanisms for the expansion, and used an expected value of information analysis to evaluate potential management interventions. The EVPI analysis reveals there is important uncertainty impeding any response, with respect both to reducing the number of brown trout and to achieving other objectives. Some potential actions, however, are more robust to uncertainty than others, pointing toward initial interventions. 3 - Coastal Sustainability Management using Avulsion Mitigation Strategies in the Yellow River Delta under Climate Change Liang Chen, Johns Hopkins University, Baltimore, MD, 21210, United States Due to high in-channel sedimentation rates, China’ Yellow River has changed course frequently, with huge socioeconomic impacts. We will address questions: What is the best timing and location of deliberate avulsions? Can use of temporary floodways lessen the cost and flooding impact, and if so where should they be located and how should they be operated? A simulation-based optimization model has been developed considering tradeoffs between cost of engineered construction and flooding risk. A robust adaptive decision model has been implemented to consider adaptive changes in operations as deep uncertainties in socioeconomic and climatic pressures evolve in the future. 4 - Value of Decision Analysis for Climate Adaptation Planning: Which Adaptation Decisions Can Benefit Most? Rui Shi, Johns Hopkins University, Baltimore, MD, United States Decision analysis, considering uncertainty and adaptability, can provide useful insights for climate adaption planning. However, adaptation managers should conduct such analyses only when the expected improvement in performance justifies the expense of the analysis. We first develop a screening tool to assess if adaptation decisions could be significantly improved by a risk-based multistage decision analysis. We then propose a procedure which quantifies nine characteristics that are associated with problems that can benefit from decision analysis. We then use the framework to prioritize adaptation decisions in the Mid- Atlantic region that are candidates for applying decision analysis. n TB42 North Bldg 227A Joint Session Analytics/Practice Curated: Analytics and Machine Learning in Healthcare Sponsored: Analytics Sponsored Session Chair: Aven Samareh 1 - Balanced Random Survival Forests for Mortality Prediction from Extremely Unbalanced, Right Censored Data Kahkashan Afrin, Texas A&M University, College Station, TX, United States, Gurudev Illangovan, Sanjay S. Srivatsa, Satish Bukkapatnam Accuracies of survival models are compromised due to the imbalance between the survival and mortality class sizes. Imbalanced datasets results in an underestimation (overestimation) of the hazard of the mortality (survival) classes. A balanced random survival forests (BRSF) model, based on training the RSF model with data generated from a synthetic minority sampling scheme is presented to address this gap. Benchmarking studies were conducted using five datasets from public repositories and a dataset of 267 patients, collected at the Heart, Artery, and Vein Center of Fresno. Investigations suggest that BRSF outperformed both Cox and RSF with an average reduction of 55% in the prediction error. 2 - Effects of Outliers and Robust Logistic Regression Tree Doowon Choi, Texas A&M, College station, TX, United States, Li Zeng This study investigates effects of outliers in logistic regression tree (LRT) modeling of binary outcome data in healthcare applications and develops a robust logistic regression tree (RLRT) approach that can alleviate those effects. The effectiveness and advantages of the proposed method is demonstrated in a case study using dataset on cardiac surgeries.

3 - A Semi-supervised AUC Optimization Model for Autism Risk Gene Prediction Ying Lin, Houston, TX, United States Autism is a constellation of neurodevelopmental presentations withhigh heritability. Prediction models, e.g. SVM and Random Forest, have been applied to automatically identified causal autism genes based on their expression patterns. But existing models are challenged by the limited training genes, imbalanced gene labels and the spatiotemporal correlations in gene expression. In this study, a semi-supervised AUC optimization model is developed to identified autism risk genes using brain specific gene expression value and their functional relationship network. It’s applied to rank more than 25,000 unknown genes and the top ranked genes are potential to enrich the autism risk genes. 4 - Med-chi: Maximum Entropy Discrimination Based Contemporaneous Health Index for Degenerative Disease Monitoring Aven Samareh, University of Washington, 4324 8th Ave NE, D7, Seattle, WA, 98105, United States, Shuai Huang We developed a novel contemporaneous health index (MED-CHI) that builds on the theory of maximum entropy discrimination. MED-CHI aims to characterize the monotonic progression characteristic underlying the longitudinal measurements for degenerative disease monitoring. Intuitively, the MED-CHI approach embodies both the data integration power of Bayesian method and the computational power of convex optimization, to handle semi-supervised structure and transfer the learned knowledge to enhance the learning ability to a completely unsupervised source domain. 5 - Dynamic Inspection of Latent Variables in State-space Systems Tianshu Feng, University of Washington, Seattle, WA, 98195, United States, Xiaoning Qian, Kaibo Liu, Shuai Huang The state-space models are widely used in a variety of areas where a set of observable variables is used to track latent variables. While most existing works focus on the statistical inferences of the latent variables based entirely on the observable variables, it comes to our awareness that in many applications, the latent variables can be occasionally acquired to enhance the monitoring of the state-space system. In our work, novel dynamic inspection methods under a general framework of state-space models are developed to identify and inspect the latent variables that are most uncertain. n TB43 North Bldg 227B Planning and Operation of Storage Assets in Electric Networks Emerging Topic: Energy and Climate Emerging Topic Session Chair: Mahdi Kefayati, Electric Power Engineers, Inc., Austin, TX, 78749, United States Co-Chair: Duehee Lee, KonKuk University, Seoul, Korea, Republic of 1 - A Scalable Computational Method for Security-constrained Unit Commitment with Energy Storage Edward Quarm, The University of Texas at Arlington, Arlington, TX, United States, Fariba Zohrizadeh, Ramtin Madani We introduce a computational method for finding feasible and near-globally optimal solutions for detailed security constrained unit commitment (SCUC) problems in the presence of energy storage units, uncertainties, and reserve requirements. The proposed method relies on conic optimization for solving mixed-integer nonlinear programming problems with large number of binary variables. Experimental results on benchmark test cases demonstrate superiority of the proposed method over off-the-shelf solvers for unit commitment. 2 - Energy Storage Planning in Presence of Topology Control Mostafa Sahraei-Ardakani, University of Utah, Salt Lake City, UT, 84112, United States, Yuanrui Sang Energy storage has the potential to alleviate the intermittency of renewable generation. One important factor in energy storage planning is the congestion patterns, which will affect the size and location of such facilities. The congestion patterns, however, are affected by topology control. This talk discusses how frequent utilization of topology control will influence the optimal location and size of energy storage facilities.

289