Covid-19: Comparison of Time Series Forecasting Models and Hybrid ARIMA-ANN

Abstract

The COVID-19 outbreak has taken many countless lives in the society and presents an unprecedented public health threat, as well as a threat to food systems and workplace safety. As a result, various machine learning models that can forecast the outbreak internationally have been developed. In order to help in the restriction of the transmission and growing number of covid cases in India, this study employs a range of methodologies to forecast the total number of current cases in India over the following 15 days. To forecast the future, the ARIMA Model, Facebook Prophet and Holt's Winter Model are utilised. Before being preprocessed, data is gathered in real time from a variety of sources. After that, the data set is split into two sections: training and testing. Finally, the model's accuracy is trained and evaluated. The approaches’ forecasting effectiveness is influenced by strategies for deconstructing the original data and merging linear and nonlinear models during the hybridization process, according to the results. By applying correct techniques, a hybrid approach can be an useful approach for enhancing forecasting accuracy achieved by traditional hybrid approaches as well as any of the other constituent methods employed alone. This paper provides a hybrid methodology that incorporates both ARIMA and ANN models to take advantage of the unique characteristics of ARIMA and ANN models in linear and nonlinear modelling. The Hybrid model showed better accuracy and a root mean square error of 21,267 which is the lowest among other models compared. The article emphasizes on the ability to assist governments in acting and making sound decisions, as well as planning for the future, in order to reduce public concern and prepare people's thinking for the next phases of the pandemic.