Comparative Analysis of Time Series Models on COVID-19 Predictions

Abstract

Many Research papers for Covid-19 prediction have been written, where researchers used different models to predict future cases. So, the objective of this paper is to perform a comparative study on all the major models and validate the results obtained before. The analysis will be performed on Indian and American Dataset. The evaluation of all the models will be performed using RMS and r <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> error. The forecast models used are ARIMA (Autoregressive integrated moving average), SARIMAX (Seasonal Auto-Regressive Integrated Moving Average with exogenous factors), and Recurrent Neural Network-based LSTM (Long Short-Term Memory) variants like Standard LSTM, Stacked LSTM, Bi-directional LSTM, Convolutional LSTM, GRU (Gated recurrent units) LSTM, and Attention LSTM. These predictive models can offer a crucial insight to policymakers and help normal citizens to prepare accordingly. Among all the mentioned models, GRU LSTM performed the best with a r <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> score of 0.986024 followed by Bi-LSTM, Attention LSTM and Stacked LSTM. Furthermore, this research study has also performed the analysis using a multivariate stacked LSTM model which outperformed all the univariate models.