A Spatio-Temporal Approach For Determining Individual's Covid-19 Risks

Abstract

Current state of art approaches such as the susceptible-infected-removed model and machine learning models are not optimized for modeling the risks of individuals and modeling the effects of local restrictions. To improve the drawback of these approaches, the feedback processing framework is proposed where previously accumulated global statistics and the model estimates generated from the spatial-temporal data are combined to improve the performance of the local prediction. The proposed framework is evaluated in three processing stages: generation of the simulation dataset, feedback analysis, and evaluation for the spatial-temporal and real-time pandemic analysis. In the data generation stage, the corresponding state of the illness for each person is modeled by a Markov stochastic process. In this stage, the parameters such as the reproduction rate, symptomatic rate, asymptomatic rate, population count, infected count, and the average mobility rate are used to update the individual's Covid-19 status and the individual's movements. The movement data of each person is generated randomly for several places of interest. In the feedback analysis stage, both the aggregated statistics and the local event data are combined in a linear model to infer a score for the Covid-19 probability of the person. In this respect, a stochastic model can be used to approximate the local statistics. In the evaluation stage, the result of the feedback analysis for all the interactions is used to classify the state of the individuals periodically. Later the accuracy of the evaluation for each person is obtained by comparing the individual's prediction with the real data generated in the same time interval. The Kappa scores independent from different populations, locations, and mobility rates obtained for every interaction indicate a significant difference from the random statistics.