Time Series Analysis using Deep LSTM Networks for predicting COVID-19 Cases in India

Shubhnesh Kumar Goyal

Abstract: While COVID has taken a toll globally, its imperative damage in India has been serious because of a large population base. In this scenario, daily prediction of COVID cases can help concerned authorities better brace themselves of the upcoming effect. Since cases form a time series data, their prediction remains a challenge due to inherent order in data points, which is tough to capture in statistical regressions. In addition, number of cases depends on a numerous factor in practical life, and to arrive on an exhaustive list for the purpose of modelling poses another challenge. To tackle these problems, we present a study spanning January 2020-April 2020, outlining way of using LSTMs for predicting 1-3 days in advance the number of cases in India and present a comparative analysis over inclusion of different factors in the prediction and its effect on accuracies. We achieved a R2 score of over 0.9 for short periods spanning 1-5 days, but model fails to capture long term (over 15 days) trend. Similarly, adding cases from Top 5 states as input factors increased the accuracy significantly for lookback = 4 to 0.99.

Keywords: Deep Neural Networks, LSTM, Epidemiology, COVID-19 prediction

How to Cite?: Shubhnesh Kumar Goyal, "Time Series Analysis using Deep LSTM Networks for predicting COVID-19 Cases in India", Volume 10 Issue 8, August 2021, International Journal of Science and Research (IJSR), Pages: 301-305, https://www.ijsr.net/getabstract.php?paperid=SR21806211654, DOI: https://dx.doi.org/10.21275/SR21806211654

Download Citation: APA | MLA | BibTeX | EndNote | RefMan