Concurrent Fusion LSTM Model for Accurate Prediction of Covid-19 Progression Using Multi-Source Data

It has become a challenge to predict the progress of COVID-19 quickly and well. Such models are good for comparing the actuals versus the predicted cases in an epidemiology study but fail in capturing most of the complex, non-linear relationships in data that end up leading to increased errors in prediction. Concurrent Fusion Long Short-Term Memory (CF-LSTM) model was developed in this study for predicting COVID-19 cases so as to address the aforementioned shortcomings in the currently available models. Multi-source real-time data such as epidemiological records, mobility, and vaccines will be coupled for better prediction. Concurrent Fusion Long Short-Term Memory model process a concurrent fusion framework that provides multifeatured input to realize spatiotemporal dependency and interaction coverage. For the public pandemic's data repository on COVID-19 by Johns Hopkins University, the performance of the models has been experimentally evaluated with the achieve worth of mean absolute error: 0.89 and root mean square error: 1.14 in respect to daily infect rates modeling performance. Scenario simulation shows promising results. It provides insights that policy makers can use in developing their intervention and resource allocation strategies. The applied model thus opens up possibilities in the application of hybrid deep learning techniques in taking prediction and management of epidemics to new levels.