Covid-19 prediction using spatio temporal recurrent neural network

Abstract

Several studies of Graph Neural Networks has lead to a better prediction of the Covid-19 evolution than many other models .Out of Many such algorithms/models LSTM(Long Short Term Memory) a time series prediction tool ,is one of them.In our work ,we make an experiment by collaborating Graph neural network and Graph Convolutional Long Short Term memory.We achieve this combination of two modules by applying Spectral Graph Convolutional operator in place of linear transformation of Long-Short_Trerm Memory module gates.We hope to exploit spatial pattern in data by this module integration.Moreover we introduce the notion of skip connection to achieve a significant improvement in jointly capturing the spatio-temporal pattern in our raw input data.We select a timing window of nearly five hundread days days from WHO-COVID-19 DATA and choose thirty Countries to train test and validate our new-cases prediction on COVID-19 .Further we test our model based on multiple error metrics like RMSE,MASE,MAE ,R2,MAPE and display their performance in tabular form.This research area has a potential in analyzing pandemic resource control ,spread forecasting and policy making application.