INFORMS 2021 Program Book

INFORMS Anaheim 2021

POSTER

12 - Nonlinear Binary Classification with Imbalanced Dataset Using Active Learning Based on Nonparametric Logistic Regression Wonjae Lee, University of Missouri, Columbia, MO, United States, Kangwon Seo The imbalance problem in a dataset is ubiquitous and inherent in data science which causes serious bias in a predictive model. It is also doubtful that the true function of classification is actually linear in covariates. This research proposes a novel data-level technique using an active learning (AL) scheme with nonparametric logistic regression to address the imbalance problem considering the nonlinear decision boundary. The preliminary experiment shows that the downsampling strategy using AL with a nonparametric model provides better performance compared to the random downsampling. 13 - Extracting the Collective Wisdom of Experts in Probabilistic Judgments Cem Peker, Erasmus University Rotterdam, Rotterdam, Netherlands How should we combine disagreeing expert judgments on the likelihood of an event? Despite its intuitive appeal, simple averaging produces an inconsistent estimator when experts have shared information. This paper proposes a novel Bayesian aggregation algorithm where experts are asked to report a probabilistic prediction and a meta-prediction. The latter is an estimate on the average of other experts’ predictions. Three experimental studies suggest that the Surprising Overshoot algorithm consistently outperforms simple averaging. Furthermore, the algorithm compares favorably to alternative aggregation algorithms in questions where experts disagree greatly. 14 - Predicting Scan Quality: A Comparison of Machine Learning Models Neda Sayahi, Wayne State University, Detroit, MI, United States, Jeremy Lewis Rickli As a relatively new technology in manufacturing metrology, X-Ray computed tomography has recently become more established. However, setting scan parameters in a quick and proper manner is challenging due to high operator dependency and lack of traceability. We argue that machine learning (ML) can accelerate parameter setting process by eliminating the need for manual setting. In this work, the accuracy of four ML methods on predicting scan quality (whether the scan will be feasible or infeasible), given a set of parameters, are compared. The results indicated that multi-layer perceptron predicted the quality of scan with high accuracy and outperformed the other methods. 15 - LP-based Characterizations of Solvable Cases of the Quadratic Assignment Problem Peter Liu, Bucknell University, Lewisburg, PA, United States, Swarup Dhar, Lucas Waddell The quadratic assignment problem (QAP) is perhaps the most widely studied nonlinear combinatorial optimization program. It boasts many applications in a variety of fields but is notoriously difficult to solve. Due to this difficulty, researchers have sought to identify special objective function structures for which the QAP is in fact readily solvable. We explain several such seemingly unrelated solvable cases in terms of the continuous relaxations of various mixed-integer linear reformulations of the QAP that are derived using the reformulation- linearization technique (RLT). 16- Eigen-entropy: A Metric for Sampling Decision Jiajing Huang, Arizona State University, Tempe, AZ, United States Hyunsoo Yoon, Ojas Pradhan, Teresa Wu, Jin Wen, Zheng O’Neill Sampling is to identify a representative data subset capturing characteristics of the whole dataset. Existing sampling algorithms have some limitations including required assumptions on data distributions or models. In this study, a new metric, termed Eigen-Entropy, is proposed, derived based on eigenvalues extracted from correlation coefficient matrix on multivariate data. The performance of the proposed method is evaluated using real building case studies. Evaluation results indicate that the proposed method outperforms the methods from existing literature in terms of accuracy while maintaining smaller number of samples. 17- Or-net: An Efficient Network for Solving Integer Programs with Deep Learning Ashton C. Kappelman, Kansas State University, Manhattan, KS, United States, Ashesh K. Sinha A new neural network architecture (OR-Net) is introduced for solving integer linear programs efficiently. This network focuses on building connections that explore the orthogonal relationships between an integer program’s coefficients. We outline implementation techniques for this OR-Net and apply it to a common knapsack problem utilizing a deep reinforcement learning framework.

5 - Drivers of Continuous Improvement Effectiveness During Covid-19: Evidence from the Nigerian Healthcare System Bukola Bakare, Western Carolina University, Cullowhee, NC, United States, Marco Lam, Olawale Durosimi-Etti, Fuad Hassan The global pandemic has taxed our modern-day health system in an unforeseen way. High demand for healthcare on already reduced resources, plus an economic downturn, is a recipe for a healthcare catastrophe in a developing country like Nigeria. As such, the implementation of continuous improvement initiatives is more important than ever. An open question then remains: how are healthcare frontline workers getting continuous improvement projects done in an extremely constrained space? This research addresses this question by investigating whether using a highly effective approach or building good relationships with employees is conducive to the success of total quality management initiatives. 6 - Forecasting the Short-Term Electric Load of Electric Reliability Council of Texas (ERCOT) Zones Using LSTM Based Deep Learning Networks Yue Wang, Texas A&M University, College Station, TX, United States, Pouya Shojaei, Jayeon Kim In this work, Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) deep learning networks are applied to forecast the electricity demand in Electric Reliability Council of Texas (ERCOT) zones for a specific week. The regional electricity consumption trend is captured by RNN and the main predictors are extracted from the given time series historical electric load and weather data, which are then trained sequentially using LSTM networks. The forecasting performance of the proposed approach is evaluated with respect to the real power consumption data in the forecasted week. 7- A Data Science Approach on Covid19 Spread Countermeasures Hamidreza Ahady Dolatsara, Assistant Professor, Clark University, Worcester, MA, United States, Gelareh Ahadi Dolatsara, Reza Poormajidi, Masoumeh Ghasemi Pirbalouti This study employs a state-of-the-art data science approach to investigate factors contributing to the spread of COVID19. Then develops an Artificial Intelligence platform to facilitate a complex decision-making process for providing an efficient countermeasure. 8 - Analytical Lessons Learned From Covid19 Data Driven Researches Hamidreza Ahady Dolatsara, Clark University, Worcester, MA, United States, Maryam Ahmadi This study reviews recent analytical researches that employed Artificial Intelligence for investigating COVID19 data. These researches are mainly related to identifying factors associated to better health outcomes in both patient and society levels, and predicting a future status based of the recorded data. The outcomes of this study help medical practitioners to employ the right analytical tools and make more efficient decisions. More specifically in the countries like Iran which per capita COVID19 cases are high and the health budget is tight. Therefore, employing the most efficient practices that backed up with Artificial Intelligence could save many lives. 9 - Google Employee Seyedali Nojabaei, Google Company, Kuala Lumpur, Malaysia Scheduling aims to enhance the correlation between healthcare resources (doctors, nurses, rooms, equipment, medicines, procedures, and management) with patient recovery and transitions after hospitalization. This processes the availability of resources, forecasting future demands for service and automating the allocation of resources to requirements. The use of artificial intelligence in scheduling makes an efficient application of the capacity. Performance and reliability are becoming major aspects in the healthcare. Scheduling plays a significant role in maintaining it. To evaluate the proposed method, a hospital case study has been conducted to show the improvement of performance. 10 - Optimal Character Selection in DND Michael A. Perry, Fresno State University, Fresno, CA, United States, Aaron Bradley Hoskins The research uses a Monte Carlo simulation to determine character survival rate in a typical one day of adventuring in Dungeons and Dragons. The Duelist Algorithm is used as an outer loop to optimize the survival rate of the adventuring party. Comparisons to other metaheuristics are also provided. 11 - Irrational Exuberance on the Crowdfunding Studies the Effect of the Covid-19 Pandemic and Government Intervention on the For reasons no one can explain, crowdfunding is surging during the pandemic with strong support from backers. We conducted this research to study the effect of covid-19 and government intervention on project supply and backer demand of crowdfunding. By analyzing the data from Kickstarter, we detected that while COVID negatively influences project supply, backers’ support increases due to COVID-evoked empathy. Our findings inform entrepreneurs to make optimal fundraising decisions, and advise crowdfunding platforms and policymakers on facilitating small-business financing, especially during economic downturns. Project Supply and Backer Demand of Kickstarter. Dan Liu, Florida State University, Tallahassee, FL, United States Guangzhi Shang, Cynthia fan Yang

30