A Distributed Ensemble of Diverse Deep Learning Models for Predicting COVID-19 Cases

Abstract

The outbreak of the COVID-19 pandemic has resulted in a significant impact on global health and economy. Forecasting the spread of COVID-19 cases is essential for policymakers to make informed decisions and allocate resources accordingly. Deep learning has shown promising results in predicting the spread of infectious diseases, including COVID-19. However, selecting the most appropriate model for a given forecasting task can be challenging, as no single model is guaranteed to perform well for all scenarios. In this paper, we propose a distributed ensemble approach that combines the predictions of different deep learning models, including Autoregressive Integrated Moving Average (Auto-ARIMA), Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM), to improve the accuracy of COVID-19 case forecasts. Specifically, we train each deep learning model on a subset of the data and aggregate their predictions to obtain an ensemble forecast. To achieve a distributed ensemble, we use an edge computing framework to train the deep learning models on multiple nodes in a parallel and distributed manner. We evaluate our approach on publicly available COVID-19 datasets, and our results show that the proposed ensemble model outperforms the individual models in terms of forecasting accuracy. Furthermore, our distributed ensemble approach significantly reduces the training and prediction time compared to training each model separately.