Predicting COVID-19 Spread using Simple Time-Series Statistical Models

Badriya Khayyat, Fouzi Harrou, Ying Sun

Research output: Chapter in Book/Report/Conference proceedingConference contribution


Accurate and timely forecasts of new COVID-19 cases and recoveries would assist in the management of medical resources and bolster public policy formulation during the current pandemic. This study aims to forecast records of confirmed time-series data using simple time series models. Importantly, to predict COVID-19 data of limited size, the performance of statistical time series models, including Linear Regression (LR) and Exponential Smoothing (ES), was investigated. The daily records of confirmed and recovered cases from Saudi Arabia, India, and France were adopted to train and test the investigated models. The forecasting accuracy has been assessed based on three commonly used statistical indicators. Results reveal that the LR model did not forecast COVID-19 time-series data successfully. On the other hand, the ES model showed a promising forecasting performance for both recovered and confirmed times-series data. Furthermore, results showed that ES outperformed the Decision Tree regression and support vector regression with linear kernel.
Original languageEnglish (US)
Title of host publication2021 International Conference on ICT for Smart Society (ICISS)
ISBN (Print)978-1-6654-1698-6
StatePublished - 2021

Bibliographical note

KAUST Repository Item: Exported on 2021-09-16


Dive into the research topics of 'Predicting COVID-19 Spread using Simple Time-Series Statistical Models'. Together they form a unique fingerprint.

Cite this