INFORMS 2021 Program Book

INFORMS Anaheim 2021

TC11

2 - A Transformer-based Approach for Soybean Yield Prediction Using Time-series Images Luning Bi, Iowa State University, Ames, IA, United States, Guiping Hu Accurate yield estimation techniques which can provide information for management decision-making is of critical importance in precision agriculture. However, traditional manual inspection and calculation is often laborious and time-consuming. To overcome the shortcomings, this paper proposes a transformer-based approach for yield prediction using early-stage images. First, a vision transformer (ViT) base model is designed to extract features from the images. Then another transformer-based model is established to predict the yield using the time-series features. A case study has been conducted using a dataset that was collected during 2020 soybean-growing seasons in Canada. The experiment results show that compared to non-time series prediction and other baseline models, the proposed approach can reduce the mean squared error by 25%-40%. 3 - Image-based Plant Phenotyping Using Deep Learning Methods Saeed Khaki, Iowa State University, Ames, IA, 50010, United States According to the United Nations, by 2050 we will need to produce 60% more food to feed the world due to global population growth. As such, agriculture which is at the heart of the food systems requires more data-driven approaches to further increase productivity, optimize management of resources, improve crop quality and quantity in a changing climate. However, the success of data-driven approaches relies on accurate and efficient collection of data. For a commercial organization that manages large amounts of crops, collecting accurate and consistent phenotypic data is a bottleneck. In this presentation, we present a state-of-the-art deep learning based method for image-based plant phenotyping which shows promise in mitigating this data collection bottleneck and fast decision-making in agriculture. TC11 CC Room 304C In Person: Analytics to Address Opioid Use Disorder General Session Chair: Md Saiful Islam, Northeastern University, Quincy, MA, 02170, United States 1 - Analyzing Long-term Health Outcomes of Patients Following the Treatment for Opioid Use Disorder: A Study of Massachusetts Commercially Insured Population Md Mahmudul Hasan, Northeastern University, Boston, MA, United States, Jiesheng Shi, Noor E. Alam, Gary J. Young We will present the findings from our investigation on patients’ long-term outcomes from buprenorphine/naloxone treatment for opioid use disorder (OUD) for a commercially insured population during a follow-up period of up to three years. The outcome of this research will potentially inform efforts that should be undertaken to increase awareness among prescribers and patients of the potential clinical value of longer treatment periods for buprenorphine/naloxone.

2 - Evaluating Policy with Matching Method From High-dimensional Data to Address Opioid Epidemic Md Saiful Islam, Northeastern University, Boston, MA, United States, Noor E. Alam Existing causal inference methods include all the observed variables just to ensure there is no confounding. Recent research shows that this approach increases the variance of the causal effect estimates without reducing the bias. In this research, we develop a novel variable selection technique to make causal inference from high-dimensional observational data. We will use all pair claim data to evaluate important policy related to Opioid Epidemic. TC12 CC Room 304D In Person: Human-Algorithm Interactions in Operations General Session Chair: Clare Snyder, Ann Arbor, MI, 48105, United States 1 - Learning Best Practices: Can Machine Learning Improve Human Decision-Making? Park Sinchaisri, Assistant Professor, Haas School of Business, UC Berkeley, Jon M. Huntsman Hall, Berkeley, CA, 19104, United States, Hamsa Sridhar Bastani, Osbert Bastani Workers spend a significant amount of time learning how to make good decisions. Evaluating the efficacy of a given decision, however, is quite complicated as decision outcomes are often long-term and relate to the original decision in complex ways. We propose a novel machine-learning algorithm for extracting “best practices” from trace data and inferring interpretable tips that can help workers improve their performance in sequential decision-making tasks. To validate our approach, we design a virtual kitchen-management game in which participants learn to minimize service time. Our experiments show that the tips generated by our algorithm are effective at improving performance, significantly outperform tips generated by human experts and a baseline algorithm, and successfully help participants build on their own experience to discover additional strategies. 2 - Does Algorithm Aversion Exist in the Field? An Empirical Analysis of Algorithm Use Determinants in Diabetes Self-management Wilson Lin, University of Southern California, Arcadia, CA, United States, Jordan D. Tong, Song-Hee Kim Advancements in algorithms hold promise to better operations by improving users’ decision-making. However, humans may exhibit so-called “algorithm aversion,” which would be a barrier to achieving such improvements — though these claims are based primarily on laboratory experiments. Using the decision- support algorithm behavior in over 230,000 bolus insulin dosing decisions from diabetes self-management, we contribute field analysis to identify drivers of algorithm use. We precisely define dynamic algorithm aversion — an asymmetric usage response to performance feedback that favors humans over algorithms — as one key hypothesis from the experimental literature. We then reject this hypothesis, instead finding that patients respond to performance feedback

asymmetrically, but in favor of the algorithm. 3 - Algorithm Use in the Service Setting

Clare Snyder, University of Michigan, Ann Arbor, MI, United States, Samantha Keppler, Stephen Leider

Computer algorithms can improve human recommendations by providing accurate advice quickly. Prior research shows however that they often do not, because people are reluctant to accept algorithms’ suggestions when recommendation quality alone is incentivized. We experimentally study the use of algorithms in a service context where servers have incentives for both recommendation quality and service time. We hypothesize that high time pressure (represented by higher customer arrival rates) will induce subjects to follow the algorithm’s recommendation. We find that under higher arrival rates, people are indeed more likely to follow the algorithm’s recommendation, especially with experience.

114