TÉCNICAS DE OTIMIZAÇÃO E MÉTRICAS DE AVALIAÇÃO APLICADAS A MACHINE LEARNING

Abstract

For machine learning, the quality of the data, as well as its pre-processing, since the algorithms used mainly derive knowledge from these same data, is decisive for the quality of the results. The presence or absence of unknown values ​​and the selection of key data, so as not to contaminate the database used with information unrelated to the objective, for example, are decisive factors when it comes to the cited merit. In many applications it is important to think about how to proceed with available information if it is incomplete as the handling of unknown values must be carefully planned or the risk of misstatement arises. This work presents a panoramic view of the subject as well as the implementation of pre-processing techniques applied to a database that contains Covid 19 data, demonstrating the processes followed and the results acquired, and showing how the pre-processing of data can influence the final result of implementing an algorithm. Learning to work with unbalanced data is essential, because in the various fields in which they are found, it is difficult to find a perfectly balanced base, therefore, understanding how to perform a good pre-processing is a possible solution to improve the results of projects with learning from machine.