Machine learning predictions of COVID-19 second wave end-times in Indian states

TL;DR

This paper uses machine learning to predict the end-times of COVID-19 waves in Indian states by analyzing SIR model predictability and optimizing parameters for accurate forecasts.

Contribution

It introduces a method combining SIR model analysis with supervised learning to improve predictions of COVID-19 wave durations in Indian states.

Findings

Optimal conditions for minimal prediction error identified

Supervised learning improves accuracy of wave end-time estimates

Method successfully applied to second wave in India

Abstract

The estimate of the remaining time of an ongoing wave of epidemic spreading is a critical issue. Due to the variations of a wide range of parameters in an epidemic, for simple models such as Susceptible-Infected-Removed (SIR) model, it is difficult to estimate such a time scale. On the other hand, multidimensional data with a large set attributes are precisely what one can use in statistical learning algorithms to make predictions. Here we show, how the predictability of the SIR model changes with various parameters using a supervised learning algorithm. We then estimate the condition in which the model gives the least error in predicting the duration of the first wave of the COVID-19 pandemic in different states in India. Finally, we use the SIR model with the above mentioned optimal conditions to generate a training data set and use it in the supervised learning algorithm to estimate the end-time of the ongoing second wave of the pandemic in different states in India.