International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 01 | Jan 2024
p-ISSN: 2395-0072
www.irjet.net
Predicting COVID-19 Cases with Time Series Analysis: An LSTM Model with Multi-Feature Integration Yaswanth Battineedi1, Srija Chaturvedula2 1Masters in Computer Science University of Florida, Gainesville, United States, yaswanth7144@gmail.
2Masters In Computer Science University of Florida, Gainesville, United States, chaturvedulasrija@gmail.com
---------------------------------------------------------------------***--------------------------------------------------------------------Key Words: COVID-19, hypertuning, LSTM, MAPE
Abstract - Accurately predicting COVID-19 cases remains
crucial for informing public health interventions and mitigating the ongoing pandemic. This study investigates the effectiveness of Long Short-Term Memory (LSTM) models for case forecasting, incorporating both traditional case data and additional features. We present a novel approach that leverages vaccination rates alongside cases to enhance prediction accuracy, while also exploring the potential impact of death data.
1.INTRODUCTION The COVID-19 pandemic has presented an unprecedented challenge to global health, economies, and societies[1]. Accurate prediction of COVID-19 cases is crucial for managing healthcare resources, implementing effective public health measures, and ultimately controlling the spread of the virus. This project addresses this issue by leveraging Long Short-Term Memory (LSTM) models, a type of recurrent neural network well-suited for time series prediction[2], to predict COVID-19 cases in the United States.
Current forecasting methods often rely solely on case data, neglecting the influence of vaccination campaigns on case dynamics. This limited approach can hinder accuracy and fail to capture the full picture of the pandemic’s progression. To address this, we propose a multi-feature LSTM model that integrates cases and vaccinations to improve prediction performance.
Existing methods for predicting COVID-19 cases primarily use case data alone[3]. These methods, while valuable, may not fully capture the complex dynamics of the pandemic. For instance, vaccination rates can significantly influence the number of future cases, and the number of deaths can reflect the severity of the disease spread. Therefore, incorporating such data can potentially enhance prediction accuracy.
Our experimental evaluation on historical COVID-19 data reveals interesting findings. Non-hyperparameter-tuned models demonstrate that incorporating vaccinations significantly improves accuracy. Model 2 (cases and vaccinations) achieves the highest MAPE of 1.24%, compared to Model 1 (cases only) with a MAPE of 5.18%. Surprisingly, including death data in Model 3 leads to a decrease in accuracy across all epoch configurations, suggesting further research into its role in prediction models.
This project introduces a novel approach to predicting COVID-19 cases in the United States. Three distinct LSTM models were developed, each using a different combination of data: cases only, cases and vaccinations, and cases, vaccinations, and deaths. This approach allows for a more comprehensive understanding of the factors influencing COVID-19 case numbers.
Hyperparameter tuning further enhances model performance. Model 2 (cases and vaccinations) maintains its superior performance with a MAPE of 1.05% after hypertuning, solidifying the benefits of incorporating multiple relevant features. However, Model 3 (cases, vaccinations, and deaths) still exhibits lower accuracy (MAPE of 8.54%) despite optimization, highlighting the need for further investigation into the optimal feature combination for accurate forecasting.
Furthermore, hyperparameter tuning was performed for each model to optimize performance. This process involved adjusting parameters such as the number of units in the LSTM layer, the activation function, and the learning rate. This fine-tuning of model parameters contributed to the improved prediction accuracy of the models.
These findings suggest that incorporating vaccinations into LSTM models significantly improves COVID-19 case forecasting accuracy. However, the inclusion of death data requires careful consideration and further research to optimize its predictive contribution. Future work could explore additional data sources, ensemble models, and alternative feature combinations to further enhance forecasting performance and inform effective pandemic response strategies.
© 2024, IRJET
|
Impact Factor value: 8.226
The technical contributions of this project include the development of LSTM models using various sets of data, the application of hyperparameter tuning to optimize model performance, and the comprehensive evaluation of the models’ prediction accuracy. The findings from this project contribute to the ongoing efforts to predict and manage COVID-19 cases effectively, providing valuable insights for future research in this area.
|
ISO 9001:2008 Certified Journal
|
Page 191