Purpose – This article compared machine learning algorithms in the context of satisfaction with democracy in Brazil. The models were trained with data from the Latinobarómetro survey, a private non-profit institution.
Theoretical foundation – The Support Vector Classifier (SVC), Random Forest (RF), and Artificial Neural Networks (ANN) classification techniques were described, followed by evaluation metrics, such as accuracy, precision, recall, f1-score and the area under a receiver operating characteristic (auc-roc).
Methodology – The data set was cleaned and the questionnaire was reduced to variables related to the local democracy index (IDL). Then, attribute transformations were performed, hyperparameters were analyzed and subsets with different class balances were created to evaluate the performance of the classifiers in different scenarios. Also, the attributes that most contributed to satisfaction with democracy were analyzed.
Results – The best classifier was RF for the class of those dissatisfied with democracy, however, the ANN and SVC techniques obtained better results in the class of satisfied individuals. Evaluating the most important attributes for satisfaction with democracy, it was identified that they are related to the country's economic situation and political and governmental issues.
Research implications – The models created were mainly able to identify people dissatisfied with democracy. The most important variables in this context were economy performance, government, political positioning, and democracy. This indicates directions for future studies and enables the development of strategies to change the perception of dissatisfied individuals.
Originality/Value – Exploring data on democracy from the Latinobarómetro using machine learning techniques in the context of Brazil.
Keywords: Machine learning; Democracy; Classification; Satisfaction.