No abstract

The high R-naught factor of SARS-CoV-2 has created a race against time for mankind and it necessitates rapid containment actions to control the spread. In such scenario short term accurate spatiotemporal predictions can help understanding the dynamics of the spread in a geographic region and identify hotspots. We propose an ensemble of convolutional LSTM based spatiotemporal model to forecast spread of the epidemic with high resolution and accuracy in a large geographic region. A data preparation method is proposed to convert spatial causal features into set of 2D images with or without temporal component. The model has been trained with available data for USA and Italy. It achieved 5.57% and 0.3% mean absolute percent error for total number of predicted infection cases in a 5day prediction period for USA and Italy respectively.

Classical Susceptible-Infected-Removed model with constant transmission rate and removal rate may not capture real world dynamics of epidemic due to complex influence of multiple external factors on the spread. On top of that transmission rate may vary widely in a large region due to non-stationarity of spatial features which poses difficulty in creating a global model. We modified discrete global Susceptible-Infected-Removed model by using time varying transmission rate, recovery rate and multiple spatially local models. No specific functional form of transmission rate has been assumed. We have derived the criteria for disease-free equilibrium within a specific time period. A single Convolutional LSTM model is created and trained to map multiple spatiotemporal features to transmission rate. The model achieved 8.39% mean absolute percent error in terms of cumulative infection cases in each locality in a 10-day prediction period. Local interpretations of the model using perturbation method reveals local influence of different features on transmission rate which in turn is used to generate a set of generalized global interpretations. A what-if scenario with modified recovery rate illustrates rapid dampening of the spread when forecasted with the trained model. A comparative study with current normal scenario reveals key necessary steps to reach baseline.

No abstract

No abstract