Utilização de métodos de aprendizado de máquina para estimação de escores de propensão

Abstract

Increasingly larger and more complex databases can be easily obtained and appropriate technologies for modeling massive amounts of data become increasingly necessary in order to optimize results and predictions. Machine learning techniques are gaining prominence in several areas and one of these is the analysis of propensity scores. In this work, the objective is to present and compare machine learning techniques. More specifically, the objective is to compare the classification tree and neural networks methodologies with logistic regression, a technique widely used in the analysis of propensity scores. Also, to know which benefits these machine learning techniques add more, to the detriment of models obtained via logistic regression, in the estimation of propensity scores. To achieve the objective, some applications are presented and discussed. The analysis procedure was developed with functions already implemented in Python libraries.