Browse by author
Lookup NU author(s): Melvin JoyORCiD
This work is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).
Background: Numerous mathematical models have been developed to forecast COVID-19 cases and helped to plan effectively by strengthening public health infrastructure and services. Many researchers incorporated the long-short term memory model (LSTM) but they have not clearly explained the workflow and steps involved in this model. Moreover, being relatively new, these models are not yet popular among biomedical researchers due to a lack of expertise. This paper presents such models as a tutorial for easy understanding and appropriate use. This includes Python codes and real-time data with instructions for implementation to forecast pandemics like COVID-19. Data and Methods: Daily cases in India from 1-Dec-2021 to 10-Feb-2022 and, in the UK from 1-May-2021 to 10-Feb-2022 were used to train the models. We used Convolutional-LSTM (CNN-LSTM) model and simple LSTM models to forecast COVID-19 cases. Models were validated using data from 11 to 25-Feb-2022. Results: CNN-LSTM and simple LSTM were fitted very well with R2 0.95 and 0.97 for India. The models were validated with RMSE and it was 9972.81 and 19285.57 for CNN-LSTM and the simple LSTM model. The R2 value of CNN-LSTM and simple LSTM models for UK data were 0.77 and 0.84 respectively. RMSE was 12111.95 for CNN-LSTM and 8935.75 for simple LSTM in the validation. Conclusion: Simple LSTM works better while training whereas the performance of CNN-LSTM was found to be better in validation. Therefore, it is suggested that train various models instead of sticking to one and revise them regularly as the behavior of an epidemic generally changes over time.
Author(s): Marimuthu S, Mani T, Joy M, Jeyaseelan L
Publication type: Article
Publication status: Published
Journal: Advances in Neurology and Neuroscience
Year: 2022
Volume: 5
Issue: 3
Pages: 130-137
Online publication date: 15/07/2022
Acceptance date: 07/07/2022
Date deposited: 20/06/2024
ISSN (electronic): 2690-909X
Publisher: Opast Group LLC
URL: https://doi.org/10.33140/AN.05.03.02
DOI: 10.33140/AN.05.03.02
Altmetrics provided by Altmetric
