Spatial-temporal prediction of high-risk areas of Covid-19 disease using Geographically Weighted Regression and Multi Layer Perceptron neural network

One of the most contagious diseases of recent years is Covid 19 disease (Corona), which has spread from Wuhan, China to the rest of the world since late 2019, causing many crises and a profound impact on the world and our daily lives. In most people infected with the disease, it causes respiratory symptoms, the severity of which depends on the person's immune system. The main objectives of this study are to discover the clusters and predict the high risk areas of Covid 19 disease, compare the efficiency of the two proposed methods and determine the effective parameters by city. In this study, Moran index and hot spot analysis index were used to investigate the distribution pattern of disease incidence rates and clusters, respectively, and Pearson correlation coefficient was used to determine effective disease parameters. In this study, statistical data of Covid 19 disease of East Azerbaijan province in the city along with environmental and topographic, health, economic and urban facilities data in the period from February 22, 2020 to November 20, 2020, were collected weekly. According to the results, the incidence of Covid 19 disease during this period has passed two peaks and according to the maps obtained from the two models, in some weeks the GWR model and in some weeks the MLP model was the superior model; also, for the GWR model, the goodness of fit index value is 0.8985 and the normalized root mean square error is 0.0822 and for the MLP model is 0.8226 and 0.1340, respectively, which shows that the GWR method is more appropriate. Sensitivity analysis of different parameters showed that the parameters of the Covid 19 incidence rate of the previous week and wind speed are more important than other modeled parameters in this issue. In this study, effective parameters were extracted separately for each city and a local model was presented that compared to the general state of the model, the local model had better accuracy than the general model of the MLP method.