The analysis and forecasting COVID-19 cases in the United States using Bayesian structural time series models

Abstract

In this paper, the Bayesian structural time series model (BSTS) is used to analyze and predict total confirmed cases who infected COVID-19 in the United States from February 28, 2020 through April 6, 2020 using the collect data from CDC (Center of Disease Control) in the United States. It includes variables of days, total confirmed cases, confirmed cases daily, death cases daily, and fatality rates. The author exploits the flexibility of Local Linear Trend, Seasonality, Contemporaneous covariates of dynamic coefficients in the Bayesian structural time series models. In addition, Causal Impact function in R programming is applied to analyze the model and read report of model. The results of the model show that the total confirmed cases who infected COVID-19 will be still most likely to increase straightly, the total numbers infected COVID-19 would be broken through 600,000 in the United States in near future (in the subsequent months). And then arrive at the peak around mid-May 2020. Also, the model suggests that the probability of variable Recovered cases daily is 0.07.

KEYWORDS:

• BSTS
• COVID-19
• causal impact
• trend seasonality
• forecast

Acknowledges

The author many thanks anonymous reviewers who took a lot time to review this manuscript and gave active comments.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Liming Xie

Liming Xie is a current PhD student who pursuing biostatistics, Bayesian, machine learning, deep learning, etc.