Previsão da pobreza do Estado do Ceará

Abstract

In recent decades, poverty and its determinants have persisted in being analyzed, with the main aim of understanding the scenario of the portion of the population that lives with insufficient income and even in unacceptable living conditions. Access to accurate and up-to-date data and techniques on poverty is essential for governments and policymakers to identify vulnerable areas, allowing them to obtain reliable knowledge through data science. Anticipating poverty is essential so that governments can help in preventing the armed forces of poverty and promoting the reallocation of resources. This study uses the Machine Learning technique to make poverty forecasts based on data from the last Family Organization Survey (POF) from 2017-2018, in a record for the state of Ceará. Various models are estimated (Logistic Regressão, LASSO and Linear Regressão). Of these methods, the one that has the greatest accuracy was the method that was LASSO and the logistical regressão had the greatest AUC ROC. Among the conclusions, it is possible to foresee a correctly classified poor taxa of 80.5% for the logistic model, and for LASSO 80.8%. It can be affirmed that 80% of the individuals on the basis of the test will be poor in the State of Ceará. Both the final models have similar impact variables, they are: type, number of people, number of children, wall, instruction, roof, type of situation, sex and age. This assumption is important because knowing which variables have a direct impact, it is possible to direct the investments that vary because it is known how important they are to anticipate poverty in Ceará.