Predição da satisfação em relação à democracia no Brasil considerando dados de pesquisa de opinião

Abstract

Opinion surveys, in questionnaires, are commonly applied in large groups and the area of democracy can be measured with this type of research, generating possibilities for the application of techniques in the area of data mining to obtain knowledge. In this work, a database was used in which the questionnaire sought to analyze positions on satisfaction with democracy, on the economy, social problems, trust in institutions and governments. From this questionnaire, this work made use of supervised machine learning (MA) methods seeking to classify satisfaction with democracy in Brazil. The classification methods used were Support Vector Classification (SVC), Random Forest and Artificial Neural Networks (ANN). and 408 attributes. In order to clean the database and reduce the questionnaire to more relevant data for the work, the Local Democracy Index (IDL) was used, which seeks through five divisions and subdivisions to assess the quality of democracy. After conducting the iterations, the best classifier was Random Forest, with results superior to the other applied methods and an accuracy of 81%. The two main attributes that contributed to a better understanding of the choice for being “satisfied” or “dissatisfied” with democracy were “Satisfaction with the economic situation (in general)” and “trust in the government”.