2021, Vol. 6, Issue 2, Part A
Otoi-NARIMA Model for forecast seasonality of COVID-19 waves: Case of Kenya
Author(s): Dr. Shem Otoi Sam, Ganesh P Pokhariyal, Khama Rogo and Dr. Edwardina Otieno Ndhine
Abstract: Background: Kenya has experienced three COVID-19 waves which left authorities mandated to do disease surveillance and estimate the burden of disease in a complex and uncertain environment with citizens’ trust in institutions wavering having lost jobs and incomes. The citizens’ vulnerability worsened with inability to connect to social support when each household wellbeing and financial ability came under threat causing much anxiety about the future. Mathematical modelling of the spread of disease informs surveillance, planning, budgeting, and response to save lives and livelihoods. In that regard, accuracy of predictions and forecasts is highly desirable. The length of duration of COVID-19 waves, the likely start and end dates, and the number of daily infections need to be estimated with precision. These inform and provide a window for authorities working in a holistic and integrated manner with researchers and experts to protect people especially the most vulnerable populations and communities to fully acquire WHO approved vaccines before the subsequent forecasted period of COVID-19 waves.
Method: Globally COVID-19 has serious health crisis with 134 million cases and 2.9 million deaths as of April 9, 2021. Kenya has experienced 392 days of COVID-19 with 136, 893 infections. The infections vary from county to county. Daily case infections data between March 13, 2020 and April 3, 2021 is used. The data is tested for stationarity and cointegration using ADF and Johansen Cointegration tests respectively. The normalized series is equally taken through these tests. A moving average of the Daily cases is estimated. The normalized series is superimposed on the moving averages. Then the combined series are used to construct Otoi-NARIMA model. The resulting model’s residual is tested for autocorrelation using autocorrelation function (ACF) and partial autocorrelation function (PAFC) tests. Also, validity of the model is tested using Ljung-Box test. The model is used to forecast 45 daily cased from April 4, 2021 to May 18, 2021. The forecasted results are visualized. Likely dates for end of third wave and potential beginning of fourth wave are picked from visualization and output of Otoi-NARIMA model. The results are compared with results of standard ARIMA model.
Results: The series and normalized version are I(1) stationary. Johansen Cointegration test revealed the existence of one cointegration rank between the series. Implication is that the series and superimposed normalized version would not drift apart overtime when used to estimate Otoi-NARIMA model. Whereas ACF revealed that both models show no autocorrelation PACF was inconclusive. On validity, the Ljung-Box test showed that both Otoi-NARIMA and standard ARIMA are valid, however the former is superior. The Otoi-NARIMA model has distinctly identified the seasonality of COVID-19 waves. In terms of visualization clarity Otoi-model provides restricted forecasts. There is likelihood of Kenya’s third wave beginning to decline briefly between April 29, 2021 and May 9, 2021. Based on assumption that Kenya will not have fully vaccinated 51 in 100 people the wave is likely to continue between June 2, 2021 and after July 10, 2021 or the third wave will continue up to June 26, 2021, which is the likely peak. It is recommended that Kenya should aim to vaccinate 51 in 100 people before June 2, 2021.
DOI: 10.22271/maths.2021.v6.i2a.675Pages: 48-58 | Views: 797 | Downloads: 26Download Full Article: Click HereHow to cite this article:
Dr. Shem Otoi Sam, Ganesh P Pokhariyal, Khama Rogo, Dr. Edwardina Otieno Ndhine.
Otoi-NARIMA Model for forecast seasonality of COVID-19 waves: Case of Kenya. Int J Stat Appl Math 2021;6(2):48-58. DOI:
10.22271/maths.2021.v6.i2a.675
