Machine Learning e a influencia das características infraestruturais e pedagógicas no desempenho escolar para o programa Milênio do MPBA

Abstract

Access to quality education is extremely important for the population, and several aspects can influence students' academic performance. Through the school census, it is possible to know the resources made available by teaching units. In the final stage of basic education, through an evaluation system, such as ENEM, it is possible, in addition to other objectives, to measure the quality of teaching in schools based on the performance of their respective students. Within this scenario, knowing which infrastructural and pedagogical characteristics most influence school performance presents itself as an important tool to act more assertively on the points that can improve the quality of teaching. Therefore, this work aims to select the most relevant infrastructural and pedagogical characteristics for student performance in public and private education units in the country. For this, microdata from the school census were used, as well as microdata from ENEM by school, with a time frame from 2009 to 2015. To carry out the selection of characteristics, two groups of attributes related to infrastructure and pedagogy were created, subsequently, selection methods based on filters, wrapper and embeded were used. The schools were classified into concepts ranging from A to D. Of the 389 columns present in the dataset, 44 were selected and their respective degrees of influence. The machine learning model resulting from feature selection achieved 98.9% of the performance of the model without feature selection.