Home / SmartTech / Google and Harvard release COVID-19 prediction models

Google and Harvard release COVID-19 prediction models

In collaboration with the Harvard Global Health Institute, Google today released the COVID-19 Public Forecasts, a series of models that predicts COVID-19 cases, deaths, intensive use, ventilator availability, and other U.S. metrics over the next 14 Days deliver counties and states. The models are trained on public data such as those from Johns Hopkins University, the Descartes Lab and the United States Census Bureau. Google says it will continue to be updated at Harvard under the guidance of its employees.

The public COVID-19 forecasts are intended to serve as a resource for first responders in the healthcare, public sector, and other affected organizations that are preparing for what lies ahead, Google says. They enable targeted public health testing and interventions from county to county and theoretically improve the ability of those who use them to respond to the rapidly evolving COVID-1

9 pandemic. For example, healthcare providers could include the forecast number of cases as a data point in resource planning for PPE, personnel, and planning. In the meantime, state and county health agencies could use infection prediction to inform test strategies and identify areas where there is a risk of outbreaks.

According to Google, the researchers have developed a novel approach to machine learning in time series to create public forecasts for COVID-19, which combines AI with a clever epidemiological basis. The design trains the models on public data and uses an architecture that allows researchers to dive into the relationships identified by the models and interpret why they are making certain predictions. They were also evaluated to ensure that predictions about colored people most affected by COVID-19 with disproportionately high case and death rates are not distorted or otherwise misleading.

Google COVID-19 forecasts

“We find that our models produce a significantly lower absolute error and normalized (relative) errors compared to the comparison model in predominantly African-American, Hispanic and white countries,” wrote Google researchers in a fairness analysis of the COVID-19 prediction models. “Our models are optimized for high accuracy in all US states to provide the best overall forecast for most communities.”

The public COVID-19 forecasts can be queried in BigQuery as part of the free 1 TB tier per month of service or downloaded as comma-separated value files (CSVs). They are also available on Google’s Data Studio dashboard and the National Response Portal.

All bytes processed in queries for the record are zeroed, according to Google, but data associated with the record is billed at the normal rate to prevent misuse. After September 15, queries about the forecast sets will revert to the normal Google Cloud billing rate.

The publication of the COVID-19 Public Forecasts follows the launch of Google’s COVID-19 Public Datasets program, which contains a repository of public data sets on the crisis and makes their access and analysis easier. COVID-19 Public Datasets program companies include the Johns Hopkins Center for Systems Science and Technology (JHU CSSE) data set, World Bank global health data, and OpenStreetMap data, all of which are stored free of charge in Google Cloud.

Source link