Using CNN-LSTMs and Transformer RNNs for COVID19 Impact Prediction
Vinayak Ashok Bharadi, Sujata Alegavi, Bhushankumar Nemade · 2023
The COVID-19 pandemic has had a significant effect on worldwide healthcare, economic growth, and communities all over the world. Forecasting COVID-19 cases accurately is critical for proactive decision-making, resource allocation, and the implementation of effective control measures. Time series forecasting techniques, particularly those based on deep learning models, have proven useful in predicting and understanding infectious disease spread. Long Short-Term Memory (LSTM) networks are gaining favor as highly effective models for time series forecasting due to their ability to capture long-term reliance and manage sequential data effectively. This study thoroughly investigates using LSTM networks in conjunction with CNNs to accurately forecast global COVID-19 daily case counts, focusing on India and the United States. The study tackles the difficult task of predicting COVID-19 case counts by considering multidimensional and multivariate factors. A suitable sequential dataset for prediction modeling is constructed by leveraging a comprehensive dataset encompassing the early stages of the pandemic. The current study assesses and compares LSTM networks and LSTM-CNN combinations to Transformer models with self-attention mechanisms, conducting a thorough analysis and benchmarking regarding prediction accuracy. The results consistently show that CNN-LSTM-based models outperform other models, with mean square errors ranging from 0.06 to 0.09. These findings highlight CNN-LSTM models' remarkable ability to capture the complex dynamics and spatial correlations inherent in COVID-19 data. By employing LSTM-based models and conducting extensive benchmarking against Transformer models, this study significantly contributes to COVID-19 forecasting and analysis.