Predicting the Impact of Covid-19 with Modified Epidemiological Model Using Deep Learning

Ivan Ji, Yixian Chen, Prakhar Mehrotra, Jialu You

Abstract: In this work, transmission rate and the related impact of COVID-19 have been systematically studied and predicted. We propose an improved modification of the standard Susceptible Exposed Infection and Recovered (SEIR) model factoring with policy-risk-related parameter. By modifying the SEIR epidemic model from deterministic differential equations to stochastic differential equations (SDEs), we improved the model reliability and usability using the extended Kalman Filter. With the time series of reproduction number R calculated from our modified SEIR with extended Kalman Filter, we can then predict the future transmission rate using novel deep learning approaches. Furthermore, to solidly reveal the impact of COVID-19, our work provides detailed and systematical study on two different levels of granularity: national and state-level. Hence, the framework proposed in this work can be effectively and accurately implemented on different region scales.