Machine Learning Applied to the INSS Benefit Request - Extended Analysis

Abstract

The materialization of social protection, foreseen in the Brazilian Constitution’s Social Security chapter, specifically in the scope of Welfare, occurs through the granting and maintenance of benefits to all Brazilians who need this protection This right generates a huge demand of millions of requests for annual benefits to the INSS (National Institute of Social Security), which is the operator of these services. Receiving and analyzing benefit requests, in a timely manner and with assertiveness, is complex and challenging. The volume of millions of applications for benefits annually, the diversity of benefits available, different criteria for granting and the urgency that the nature of these applications requires for the maintenance of life for applicants, express this complex and challenging environment. Within this context, the present study aims to develop some models, using machine learning techniques, and select the best one, which can predict whether a certain benefit request will be granted or rejected. This prediction would help in the analysis of new benefit requests, making room for the dynamics of the analysis process to be directed more quickly and assertively. The data source for the construction of the models in this work was obtained from the INSS Open Data Portal, which are included in the INSS Open Data Plan. This dataset is composed of monthly files of Decided Benefits (Granted and Dismissed) from December 2018 to June 2020. As a scope of analysis, algorithms such as KNN, SVC, Decision Trees, Logistic Regression, etc. were addressed. Models were also built using the Ensemble Bagging and Boosting techniques, reaching a set of seventeen analyzed algorithms. The algorithm that obtained the best performance, using the F1 metric as the determinant, was the eXtreme Gradient Boosting (XGB) Classifier with 80%. With this, the model performs the prediction with approximately 84% Accuracy, 76% Sensitivity and 81% for AUC (Area Under the Curve). As a result of the study, a model capable of predicting whether a given benefit requirement would be granted or denied was obtained, based on the requirement data, with a performance within the expectations established in the objectives. Therefore, this article analyzed seventeen machine learning algorithms with the aim of building a model to predict whether a given benefit request to the INSS will be granted or rejected. As a contribution of this article to the IS, we highlight the initiative to apply a machine learning model in a domain not yet explored by other research, using, as a data source, open data made available by the federal government in the cloud. It is hoped that the result of this study will open space for new scientific research in the field of machine learning to be developed in this domain, with the aim of helping with real problems in this sensitive part of the lives of Brazilians and of public administration.