Detecção de atributos que melhor caracterizam perfis de inscritos do ENEM utilizando Redução de Dimensionalidade

Abstract

Since 2010, the Exame Nacional do Ensino Medio - ENEM has been used as the only exam to apply for a position in some Brazilian. Because of this, the number of candidates interested in the exam has increased and a large volume of data has being collected from them. This has made ENEM a high dimensional dataset. Data mining can extract knownledge from a large volume of data. One of the most important data mining technique is clustering. By using this technique it is possible to identify profiles of members based on socioeconomic questions usually answered by ENEM candidates. It is important to use feature selection method in high dimensional databases to identify the most characterizing attributes of a dataset and improve the efficiency, since high dimensional data is more costly to process. Futhermore, the random choice of attributes can produce incorrect or useless results. By using different approaches to select attributes, this study aims to compare the filter and wrapper approaches with different searching algorithms and then identify the one of the best results. In this work, the ENEM data from 2010 was used to identify profiles via clustering algorithm, trying to relate the average grade obtained in the exam with socioeconomic data provided by each candidate upon his/her registration. In addition, an analysis was made between the clusters obtained in this study and the work of (CAMINHA; MOREIRA; SILVA, 2015) which makes a similar study using the same data. This study was concluded that socioeconomic factors do not have major impact on the final grade and can not therefore influence it (at least for 2010 data).