Abstract — This work is an attempt to examine empirically the best ARIMA model for forecasting the number of infections of COVID19 in the Kingdom of Saudi Arabia. The historical data of daily infections were used for this purpose.

At first, the stationarity of the data series was observed by conducting ACF / PACF plots in addition to the statistics such as as Ljung-BoxPierce Q-statistic and Dickey-Fuller test statistic.

At the beginning, the data were found Non-Stationary as the P-Value of ADF = 0.97 and Ljung-BoxPierce P-Value was way below 0 which indicates the non-stationarity.

But after the first difference values of the data series the same types of plots and statistics show the data is stationary as ADF P-Value = 0.01 and Ljung-BoxPierce P-Value = 0.10 which was better than previous results.

The best ARIMA model have been selected by using the criteria such as AIC, MASE, RMSE, MAPE after splitting the observations into two groups A) learning data — B) testing data.

The model for which the values of criteria are best fit are considered as the best model.

Hence, ARIMA (2,0,2) is found the best model for forecasting in terms of AIC and MASE.

And ARIMA model (1,2,2) in terms of MAPE and P-value. Then using the two models forecast of the data has been made and values of the period from 26th Aug 2020–1st Sep 2020 are predicted and reported in the study.

I. Objectives of the study

Data of the daily infections in Saudi Arabia was obtained through establishing direct connection with Saudi Arabian Ministry of Health API which was provided for particular information about COVID19 daily cases.

The API providing the data accumulatively day by day that was treated into daily infections by mutating the data by adding a column that takes the difference of daily reported cases and subtracting the lead day from the next day:

mutate (lag = lead(acc_cases,1L))%>%

mutate (daily = abs(acc_cases-lag))

#arima #data-science #forecasting #covid19 #saudi-arabia #data analysis