A Time-Dependent SEIRD Model for Forecasting the Transmission Dynamics in Infectious Diseases: COVID-19 a Case Study

The spread of a disease caused by a virus can happen through human-to-human contact or from the environment. An estimation is crucial to make policy decisions and plan for the medical emergencies that may arise. Many mathematical models extend the standard SIR model to capture disease spread and estimate the infections, recoveries, and fatalities that may result from the disease. One major factor important in the forecasts using the models is the dynamic nature of the disease spread. Unless we can develop a way to guide this dynamic spread, estimating the parameters may not give accurate forecasts. To capture the transmission dynamics, we implement a time-dependent SEIRD model. In this data-driven model, we try to estimate parameters from the equations derived from the traditional SEIRD model. The main principle is to keep the model generic while making minimal assumptions. In this work, we have derived a data-driven model from SEIRD, where we attempt to forecast infected, recovered, and deceased rates of COVID-19 for the next 21 days. A method for estimating the dynamic change in the parameters of the model is the crucial contribution of this work. The model has been tested for India at the district level and the USA at the state level. The mean absolute percentage error (MAPE) obtained for predicting confirmed/deceased for day 7 is between 4–5%, by day 14 is about 8–10%,and 12–15% for day 21. A dashboard has been developed based on the proposed model showing the predictions for active, recovered, and deaths at the district level in India [1]. We believe that these forecasts can help the governments in planning for emergencies such as ICU requirements, PPEs, and hospitalizations during the spread of infectious diseases.