Author(s): Shem Otoi Sam and Edwardina Otieno Ndhine

Abstract: In this paper we are looking for the best model to predict COVID-19 cases, fatalities, and active cases. Time series analysis using auto-regressive integrated moving averages is used. The three series are tested for stationarity using Augmented Dickey-Fuller test (ADF) and differenced to obtain stationarity. Both the data and models formulated are tested for autocorrelation using ACF and PACF. In selecting the best ARIMA model Akaike Information Criteria (AIC) that gives the least value is picked. According to AIC selection the best model for cases, fatalities, and active/infected are ARIMA (2, 2, 2), (1, 2, 3), and (0, 2, 1) respectively. It implies that two lags of previous cases have influence on current case occurrence as opposed to one lag of fatalities. Asymptomatic and infected people with clear syndromes, not in quarantine, may move and interact with others leading to more infections, compared to deceased. The residuals of estimated case fatalities and cases models are not autocorrelated as seen from the ACF and PACF. Also, the forecast of case fatality model show that fatalities will stabilize on 9/8/2020 and begin to fall from 22/8/2020.The cases model forecasts that COVID-19 cases stabilize after 17/8/2020 and begin to fall from 23/8/2020, both dates will have 35,779 and 41,517 COVID-19 total cases respectively. From further analysis, the forecast of active cases over time gives additional information; its lower limit forecast begins to fall after 10/8/2020 showing 13669 infected persons and reduces to 10345 on 4/9/2020. Consequently, 10/8/2020 is a possible beginning of Kenyan peak. There is also a statistical possibility that the peaks of cases, fatalities, and infected persons occur at different intervals of time or rotating seasonal peaks. The progress of active cases over time carries the “energies” or “momentum” of COVID-19 pandemic.

Pages: 249-257 | Views: 671 | Downloads: 20

Download Full Article: Click Here