Utilização de agrupamento como método de pré-processamento em problemas de regressão linear

Abstract

For real-world data to be explored, pre-processing is needed in order to ease machine learning applications. Nowadays, various pre-processing techniques are available, and this paper aims to show the impacts of using clustering as one of them, more specifically in linear regression problems. Two experiments were carried out, using two different databases: the first one describing data from a series of properties from Ames, a city from Iowa, in the United States of America, and the second one containing information about COVID-19. It was observed that in both cases, clustering before applying the regression model improves regressor performance, based on database nature. Besides the improvement, it is recommended to use clustering alongside other pre-processing techniques.