Monitoring and Forecasting COVID-19: Statistical Heuristic Regression, Susceptible-Infected-Removed model and, Spatial Stochastics

TL;DR

This paper compares a simple statistical heuristic regression and a minimal SIR model for COVID-19 forecasting, demonstrating their accuracy and utility in predicting disease evolution, and explores stochastic spatial models for detailed insights.

Contribution

It introduces and benchmarks a minimal SIR model combined with SHR for COVID-19 prediction, highlighting their accuracy and practical usefulness over data-driven approaches.

Findings

SHR achieves about +/- 2% accuracy 20 days after the second inflection point.

SIR model reaches similar accuracy approximately two weeks earlier.

Stochastic spatial models provide detailed disease spread insights beyond SIR and SHR.

Abstract

The COVID-19 pandemic has had worldwide devastating effects on human lives, highlighting the need for tools to predict its development. Dynamics of such public-health threats can often be efficiently analysed through simple models that help to make quantitative timely policy decisions. We benchmark a minimal version of a Susceptible-Infected-Removed model for infectious diseases (SIR) coupled with a simple least-squares Statistical Heuristic Regression (SHR) based on a lognormal distribution. We derived the three free parameters for both models in several cases and tested them against the amount of data needed to bring accuracy in predictions. The SHR model is approximately +/- 2% accurate about 20 days past the second inflexion point in the daily curve of cases, while the SIR model reaches a similar accuracy a fortnight before. All the analyzed cases assert the utility of SHR and SIR approximants as a useful tool to forecast the evolution of the disease. Finally, we have studied simulated stochastic individual-based SIR dynamics, which yields a detailed spatial and temporal view of the disease that cannot be given by SIR or SHR methods.