EAGER: Epidemic Spread Modeling Using Hard Data

数据驱动的COVID-19传播预测模型对于指导公共卫生政策至关重要。以汇总形式作为输入数据的流行病学模型可用于预测，但输入的粒度会限制模型的可用性。以个体为中心的模型更加灵活，但需要输入群体中每个人移动的时间序列：每个人的确切位置，每个人在该位置停留的时间，以及向下一个位置的过渡。由于隐私问题，这种粒度的准确数据是不公开的。该项目的重点是开发一个以个人为中心的预测生态系统，可用于预测高度传染性疾病在城市地区活跃人群中的传播。这样的模型可用于制定减轻疾病传播的假设情景，并可成为指导未来大流行决策的不可或缺的工具。该项目将为COVID-19和未来大流行的流行病建模提供灵活的工具。该项目提倡使用基于主体的模型作为机器学习的替代方案，以准确预测传染病的传播。目的是创建一个预测生态系统，以评估详细的情景：流动性的地理限制、在家工作订单/咨询、学校关闭（以及在不同条件下的部分开放）、以不同能力运作的兴趣点、隔离时间和疫苗接种优先级等。上述场景可以在不同的细节级别上建模，目的是保持模型输入小、紧凑和灵活，但不影响其预测能力。在基于代理的模型设置中对上述内容进行分析，识别出最有效且可行的输入抽象，类似于识别机器学习模型中特征选择的重要性。该工具由匿名手机数据驱动，将提供一个强大的建模生态系统，利用随机模型捕捉传染病缓解措施的效果，这些模型与机器学习模型相辅相成。通过这个项目，本科生和研究生将在应用数据科学的艺术训练。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Data-driven prediction models of the spread of COVID-19 are critical for guiding public health policy. Epidemiological models that use as input data in aggregated form can be used for prediction but the granularity of input can limit model usability. Models that are individual-centric are a lot more flexible but require as input the time series of every person's movement within a population: the exact location of each individual, the duration of the individual's stay at the location, and the transition to the next location. Due to privacy issues, accurate data of such granularity are not publicly available. The focus of this project is on the development of a prediction ecosystem that is individual-centric and can be used to foresee the spread of a highly contagious disease within a population that is active within an urban area. Such a model can be used to develop what-if scenarios to mitigate the spread of the disease and can become an indispensable tool for guiding policy decisions in future pandemics. This project will provide a flexible tool for epidemic modeling of COVID-19 and future pandemics. This project advocates the usage of agent-based models as an alternative to machine-learning for accurate prediction of the spread of contagious diseases. The aim is to create a prediction ecosystem for evaluating detailed scenarios: geographical restrictions of mobility, work from home orders/advisories, school closures (and partial openings under different conditions), points of interest operating under various capacities, time in quarantine, and vaccination priority, among others. The above scenarios can be modeled at various levels of detail with the aim to keep the model input small, compact, and flexible, but without compromising its prediction ability. Analysis of the above within the agent-based model setting identifies the most effective yet feasible input abstractions, similar to identifying the importance of feature selection in machine learning models. This tool, driven by anonymized cell-phone data will provide a robust modeling ecosystem that captures the effect of mitigation measures of contagious diseases using stochastic models that are complementary to machine-learning ones. Through this project, undergraduate and graduate students will be trained in the art of applied data science.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.