Modelo heurístico de seleção de características em aprendizado de máquina

Abstract

Over the years, there has been a breakthrough in technology that has enabled people to generate more and more data, and the concept of Big Data has emerged. As a result, there was a need to treat them more automatically, leading to the development of techniques for the Knowledge Discovery process. Among the steps of the KDD, there is a reduction in dimensionality that is very important, because it eliminates attributes that do not contribute to the analysis. This work carried out a comparative analysis of Filter approaches already existing in the literature with a new method of Wrapper developed to reduce the dimensionality together with the classifiers Naive Bayes, KNN, AdaBoost and random forests, to verify which presented better performance in accuracy, precision and/or recall.