Time series model for forecasting the number of COVID-19 cases in Turkey

Abstract

Objective: Coronavirus disease 2019 (COVID-19) had an unprecedented effect on bothnations and health systems. Time series modeling using Auto-Regressive IntegratedMoving Averages (ARIMA) models have been used to forecast variables extensively instatistics and econometrics. We aimed to predict the total number of cases for COVID19using ARIMA models of time-series analysis in Turkey.

Methods: We used timeseries analysis to build an ARIMA model of the total number of cases from March 11,2020 to August 24, 2020 and used the model to predict cases in the following 14 days,from August 25, 2020 to September 7, 2020. Hyndman and Khandakar algorithm wasused to select components of ARIMA models. Percentage error was used to evaluateforecasting accuracy.

Results: During the model building period, 259692 cases werediagnosed and during 14 days of validation period additional 21817 new cases wereadded. ARIMA model with (p,d,q) components of (4, 2, 0) was used for forecasting.The mean percentage error of forecast was 0.20% and forecast accuracy was highestin the two weeks of forecasting.

Conclusion: ARIMA models can be used to forecastthe total number of cases of COVID-19 patients for the upcoming two weeks in Turkey

Keywords

• Covid-19
• time series
• forecasting

Türkiye’de görülen COVID-19 olgu sayılarının tahmininde zaman serisi modelinin kullanılması

Abstract

Amaç: Koronavirüs hastalığı 2019’un (Covid-19) hem ülkeler hem de sağlık sistemleriüzerinde beklenmedik bir etkisi olmuştur. Otoregresif Entegre Hareketli Ortalama(Auto-Regressive Integrated Moving Averages) (ARIMA) modellerini kullanarakyapılan zaman serisi modellemesi, istatistik ve ekonometride değişkenleri kapsamlışekilde tahminde kullanılmaktadır. Zaman serisi analizinin ARIMA modellerinikullanarak, Türkiyede Covid-19 için toplam olgu sayısını tahmin etmeyi amaçladık.

Yöntem: 11 Mart 2020’den 24 Ağustos 2020’ye kadar olan toplam olgu sayısınınbir ARIMA modelini oluşturmak için zaman serisi analizini kullandık ve 25 Ağustos2020’den 7 Eylül 2020’ye kadar takip eden 14 gündeki vakaları tahmin etmek içinbu modelden yararlandık. ARIMA modellerinin bileşenlerinin seçiminde Hyndmanve Khandakar algoritması kullanıldık. Öngörme doğruluğunu değerlendirmekiçin yüzde hata kullanıldı.

Bulgular: Model oluşturma döneminde 259.692 olguteşhis edildi ve 14 günlük doğrulama süresi boyunca ek 21.817 olgu vaka eklendi.Öngörü için (4, 2, 0) bileşenli (p, d, q) bileşenli ARIMA modeli kullanıldı. Ortalamatahmin hatası % 0.20 olarak bulundu ve tahmin doğruluğu tahminin iki haftalıkdöneminde en yüksekti.

Sonuç: ARIMA modelleri, Türkiye’de önümüzdeki iki haftaboyunca Covid-19 hastalarının toplam olgu sayısını tahmin etmek için kullanılabilir.

Keywords

• COVID-19
• zaman serisi
• tahmin

References

1. Bourouiba L. Turbulent Gas Cloudsand Respiratory Pathogen Emissions:Potential Implications for ReducingTransmission of COVID-19. JAMA. 2020,March 20, online ahead of print.
2. Kannan S, Shaik Syed Ali P, SheezaA, Hemalatha K. COVID-19 (NovelCoronavirus 2019) - Recent TrendsEur Rev Med Pharmacol Sci. 2020Feb;24(4):2006-2011.
3. Adhikari SP, Meng S, Wu YJ, Mao YP, YeRX, Wet al. Epidemiology, causes, clinicalmanifestation and diagnosis, preventionand control of coronavirus disease(COVID-19) during the early outbreakperiod: a scoping review. Infect DisPoverty. 2020;9:29.
4. Driggin E, Madhavan MV, Bikdeli B,Chuich T, Laracy J, et al. CardiovascularConsiderations for Patients, Health CareWorkers, and Health Systems During theCOVID-19 Pandemic. J Am Coll Cardiol.2020;75:2352-2371.
5. Box GEP JG, Reinsel GC. Time seriesanalysis: Forecasting and control. Delhi:Pearson Education, 1994.
6. Zhou L, Zhao P, Wu D, Cheng C, HuangH. Time series model for forecasting thenumber of new admission inpatients.BMC Med Inform Decis Mak. 2018;18:39
7. Juang WC, Huang SJ, Huang FD, ChengPW, Wann SR. Application of time seriesanalysis in modelling and forecastingemergency department visits in amedical centre in Southern Taiwan. BMJOpen. 2017;7:e018628
8. Covid-19. Web address: https://covid19.saglik.gov.tr/TR-66935/genelkoronavirus-tablosu.html# Accessed:September 10, 2020

9. Agha R, Abdall-Razak A, Crossley E,Dowlut N, Iosifidis C et al. for the STROCSSGroup. The STROCSS 2019 Guideline:Strengthening the Reporting of CohortStudies in Surgery. International Journalof Surgery 2019;72:156-165.
10. General Notation for ARIMA models,Web address: https://v8doc.sas.com/sashtml/ets/chap7/sect8.htm accessed:June 16 ,2020.
11. Hyndman, R, Khandakar Y. AutomaticTime Series Forecasting: The ForecastPackage for R. J Stat Softw 2008;27:1–22.
12. Kermack, WO, McKendrick AG. AContribution to the MathematicalTheory of Epidemics. P Roy Soc A-MathPhy. 1927;115:700–721.
13. Yang Z, Zeng Z, Wang K, Wong SS, Liang W,et al. Modified SEIR and AI prediction ofthe epidemics trend of Covid-19 in ChinaUnder Health Interventions. J Thorac Dis2020;12:165-174. d
14. Roda WC, Varughese MB, Han D, Li MY.Why is it difficult to accurately predictthe covid-19 epidemic? Infect Dis Model2020;5:271-281.
15. WHO Pandemic Influenza RiskManagement. Web address: https://www.who.int/influenza/preparedness/pandemic/influenza_risk_management_update2017/en/ Accessed June 16,2020.