A multivariate spatiotemporal spread model of COVID-19 using ensemble of ConvLSTM networks

swarna kamal paul; Saikat Jana; Parama Bhaumik

The pixel values are transformed to ∆I and summed up cumulatively to calculate total infection cases I, up till a specific day. MAPE is calculated at pixel level for total infection cases at the end of prediction period and averaged. The pixels with 0 susceptible population count are ignored while calculating MAPE. Pixel MAPE is calculated as per equation 2, where is set of all unique pixels in all grids such that the frame for each corresponding grid have non zero total infection count, ′ is prediction time period, ′′ = − ′ is total time period in training set, is a pixel from a set of unique pixels in the total region having non zero actual susceptible population, is predicted pixel value on th day, is susceptible population at pixel on th day calculated from and −1 , ∆ is actual new infection count at pixel on th day and is total number of pixels in the region. ̂ and are total predicted and actual infection cases respectively.