In this project, a multivariate deep-learning based Early Warning System (EWS) is implemented for respiratory infection outbreaks. Data is gathered from multiple online sources including Google trends, Google news, Wiki trends, Reddit, Google Earth Enginge (GEE), Open-Meteo, and underground weather. The datasets are combined and used for training, validating, and testing the models. The models are composed of four layers, namely, a Convolutional Neural Network (CNN), a Graph Neural Network (GNN), a Gated Recurrent Unit (GRU), and a stacked linear Neural Network (NN). The CNN layer combines the data sources into one vector. The GNN and GRU layers analyze the data on spatial and temporal deminsions, respectively. Finally, the stacked NN layer provides a T step-ahead prediction. The machine learning pipeline is automated to collect the datasets, train and test the models, forecast the coming waves, and visualize the results in this dashboard. The models are capable of forecasting the outbreaks with an outstanding accuracy by up to 56 day(s) (8 week(s)) in advance.

For detailed information on the datasets and methodology, please refer to our manuscript.

COVID-19 EWS

The datasets are freely available to copy, use, and redistribute for non-commercial purposes only, provided that the authors are appropriately credited.

Download full dataset for COVID-19 in African countries (.csv):

Download full dataset for COVID-19 in Canadian provinces (.csv):

Two models are trained for predicting the COVID-19 number of cases in (1) African countries, and (2) Canadian provinces. For each model a graph is formed and provided to the GNN layer. The nodes of the graph are countries for the African, and provinces for the Canadian framework. The weights of the graphs are set to the correlation of the COVID-19 number of cases between the two nodes. The number of COVID-19 cases are collected on daily basis from the WHO website for African countries and the COVID19Tracker API for Canadian provinces. The underlying data for the Canadian system includes Google trends, number of Reddit posts, and GEE air quality concentrations, and for the African system is the same, plus the number of released Google news article and the number of Wikipedia page views.

Number of COVID-19 Cases

Number of COVID-19 Cases