Comparative Performance Analysis of Machine Learning algorithms for COVID-19 cases in India

Abstract

A novel coronavirus is the cause of the viral infection recognized as COVID-19 (initially named as SARC-CoV-2). Since the pandemic emerged in the Wuhan province of China in November 2019, it has been recognized as a global threat. However, over the next two years, it has been witnessed that the novel coronavirus tends to evolve rapidly. In this paper, we leverage our time-series data collected since the initial spread of COVID-19, mainly in India, to better understand the growth of this pandemic in different regions throughout the country. The research is based on cases reported in India in chronological order. In addition to numerous previous works, we have tried to come up with the most appropriate solution to estimate and predict the newly reported COVID-19 cases in the upcoming days, with the least possible error through machine learning. This study also aims to compare multiple machine learning algorithms on various factors and their trade-of for prediction. The experimental results specify that Orthogonal Matching Pursuit is the best algorithm for this problem. We make our dataset available for further research.