Estimação da frequência de utilização de serviços de saúde durante a pandemia utilizando machine learning

Estimating the frequency of use of health services during the pandemic using machine learning

Abstract

Data mining has been a highly widespread tool since the emergence of computing, performing the collection, filtering, processing, analysis and obtaining relevant information in complex databases. Within the range of data mining applications, machine learning stands out, that is, algorithms designed in machines so that they "learn" to work with different data effectively. Machine learning techniques have been widely used for the analysis of databases of different natures, with special emphasis on those generated with data collected from the advent of the new coronavirus pandemic. In this article, three techniques were used to estimate the number of visits to health services in countries across the Americas, and later comparing each of them with Brazil. The three techniques used were random forest algorithms, neural networks and k-nearest neighbors (KNN). From the application of the three algorithms, the mean absolute error, the mean square error, and the root mean square error were then analyzed for comparison purposes. Based on the results obtained, it was observed that the country that had the most distinct population from the Brazilian population in the subject studied is the United States of America, while the two most similar countries are Uruguay and Honduras.