Avaliação de algoritmos de aprendizado de máquina e modelo explicativo para a análise da satisfação quanto à economia e seus principais fatores de contribuição

Abstract

This study aims to develop machine learning models to classify the satisfaction of people in Latin America regarding their country's economy and analyze the factors that most influence this variable. Data mining is proposed as an innovative alternative to obtain more detailed and non-trivial information about population satisfaction. The study involves exploring and preprocessing survey data provided by the Latinobarómetro for the period between 2016 and 2020, totaling 80,816 respondents across approximately 400 questions. The selected classification models are Random Forest, Gradient Boosting Classifier, XGBoost, and Naïve Bayes. These models were evaluated using precision, recall, f1-score, and accuracy metrics. To achieve better results, the learning models were combined using ensemble methods: Hard Voting and Soft Voting, with the models reaching 86% and 85% accuracy, respectively. Finally, the ensemble models underwent an explanatory analysis using the Shapley Additive Explanations (SHAP) method. Considering the complexity of socioeconomic variables in Latin America, including historical issues and challenges, understanding the factors that influence the perception of the economy is crucial for guiding actions that improve living conditions. Furthermore, this study seeks to fill the existing gap in the application of data mining in the context of economic satisfaction, contributing to the advancement of this research area.