Change Point Detection For COVID-19 Time Series In Typical Countries Around The World

The COVID-19 has attracted much attention since its outbreak in December2019 and the development in various countries and regions around the world is obviously different.Based on the existing research results,this paper combines the concept of ‘sliding window’ with the statistical analysis method of change points and applies it to the field of public health.It selects the United States,Brazil,the United Kingdom,Russia,India,Indonesia,Philippines and China as the research objects,and conducts mean change point test on the time series of daily growth rates of cumulative confirmed cases from January 3,2020 to November 1,2021.It aims to quantitatively describe the development trend of each country’s epidemic situation,analyze the reasons for the change points and conduct extended researches on multiple dimensions.The main results obtained are listed as follows:First,the simulation experiment shows that the SN-based test is more stable than the KS test,so the SN-based test is selected for the study of COVID-19 time series in this paper.Second,the location of the change point and the implementation time of the policies and measures is closely related.There is generally a certain lag effect.Third,according to the trend of the cumulative confirmed case curve,8countries can be divided into three types: long-term stable,phased fluctuation and continuous expansion.Fourth,the comparison of the detection results of the mean change point and the development situation of COVID-19 between countries shows that,the trend within 180 days from the starting position of the change point detection has a strong time correlation with the country’s first release of strict epidemic prevention measures such as home isolation and social distancing restrictions.Fifth,in order to further verify the mean change point detection results,ARMA models are selected to fit the stationary subsequence between adjacent change points.The effect of the fitting model varies from country to country,and individual countries have a high goodness of fit,which can reach more than 0.85.