Explaining Causal Influence of External Factors on Incidence Rate of Covid-19

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 and spatio-temporal variation of transmission rate. Also, explainability of a model is of prime necessity to understand the influence of multiple factors on transmission rate. Thus, we modified discrete global susceptible–infected–removed model with time-varying transmission rate, recovery rate and multiple spatially local models. We have derived the criteria for disease-free equilibrium within a specific time period. A 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 region in USA for a 10-day prediction period. Comparison with current state of the art methods reveals performance superiority of the proposed method. A perturbation-based spatio-temporal model interpretation method is proposed which generates spatio-temporal local interpretations. Global interpretations are generated by statistically accumulating the local interpretations. Global interpretations of transmission rate for a region in USA shows consistent positive influence of population density, whereas, temperature and humidity have very minor influence. An experiment with what-if scenario reveals locality specific quick identification of positive cases, rapid isolation and improving healthcare facilities are keys to rapid convergence to disease-free equilibrium. A long-term forecasting of 160 days predicts new infection cases in a region in USA with a median error of 455 cases.