Manuscript received June 8, 2023; revised July 28, 2023; accepted August 24, 2023; published March 20, 2024
Abstract—This paper explores the three variants of Long Short-Term Memory (LSTM) deep learning models for the analysis and prediction of univariate time series data to develop better understanding of the spread of COVID-19 pandemic. The COVID-19 pandemic continues to significantly impact public health, medical and industrial infrastructure, and the economy. Many researchers are working on various computational and mathematical models to analyze the underlying causes of transmission and spread of COVID-19. Accurate predictive models for COVID-19 will be significant as they will help us better manage the resources of health care professionals and medication to combat the COVID-19 pandemic. This study explores three deep learning models based on the Long Short-Term Memory (LSTM) cells to analyze and predict sequential data. These models are vanilla LSTM, bidirectional LSTM, and stacked LSTM. The models were trained and tested using the univariate time series data of daily trends in number of COVID-19 cases in the United States. Data was collected from the Centers for Disease Control and Prevention website (CDC) from January 23, 2020, to May 25, 2022. The models were trained using the first 743 samples and tested on the last 104 samples. The models were implemented using TensorFlow, and KerasTuner was used to optimize the hyperparameters of LSTM networks. The prediction accuracy of three models was compared using the metric of Mean Absolute Percentage Error (MAPE). It was found that bidirectional LSTM gave the best accuracy.
Keywords—Long Short-Term Memory (LSTM), deep learning, univariate time series, data, COVID-19, comparison of different LSTM models
Cite: Ashish Kharel, Zeinab Zarean, and Devinder Kaur, "Long Short-Term Memory (LSTM) Based Deep Learning Models for Predicting Univariate Time Series Data," International Journal of Machine Learning vol. 14, no. 1, pp. 30-37, 2024.
Copyright © 2024 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).