Identifying the Most Relevant Attributes to Explain Peaks of COVID-19 Infections and Deaths by Machine Learning Methods

Abstract

One of the key factors related to assessing the spreading speed of a given disease is to determine the peak of infections, the point after which a wave starts to mitigate as the daily number of cases goes down. This issue has attracted the attention of scientists for the last two years in relation to the COVID-19 pandemic. At the present time, since several waves have affected most countries, there is plenty of information at our disposal: date and magnitude of contagion peaks; country-related data such as population density, gdp per capita, etc.; among other relevant status metrics at the dates of peaks, like vaccination, mobility, use of mask, occupied hospital beds, etc. Thus, finding which of those attributes are relevant and ranking them becomes an interesting field for research. In this work, we apply a filtering technique to identify peaks on the reported data and then perform feature selection algorithms with the peak magnitude as output. A comparative ranking of the attributes is thus obtained for several countries and for different waves in the same country. As pre-processing tasks, we performed a normalization and a conversion from numerical to categorical values on the output variable. As a result, a grouping of countries and waves is obtained, from where important information can be extracted. Our results contribute with knowledge for predicting and monitoring the spreading of diseases and become a relevant tool for health institutions and authorities.