Machine Learning Applied in the Construction of a Disability Retirement Entry Table of the General Social Security Regime (RGPS) of Brazil
Resumo
Context: Given the significant expenses incurred by the General Social Security Regime (RGPS) in disability retirement benefit payments, accurately assessing the costs associated with new concessions is crucial for maintaining the financial and actuarial balance of the system. Entry tables for disability retirement are essential instruments for this measurement. Problem: In addition to the scarcity of related studies, the most recent table is based on RGPS concession data from just over twenty years ago, which can impact the precise estimation of costs for new concessions. Solution: This article aims to construct an updated Disability Retirement Entry Table in RGPS, segmented by gender and age, employing Machine Learning models developed using regression algorithms. IS Theory: This article operates under the auspices of Machine Learning Theory, which is related to automating intelligent tasks to identify patterns or mathematical formulations explaining potential data relationships. Method: The research utilizes a quantitative approach, involving the training of Machine Learning models through experiments with regression algorithms and performance evaluation metrics such as RMSE and R2. Summary of Results: The Extra Trees Regressor demonstrated the best performance, achieving results of 0.002039 and 0.917324 for RMSE and R2 metrics, respectively, with training data, and 0.001047 and 0.975334 with out-of-sample data. Contributions and Impact in the IS Area: This article’s primary contribution lies in pioneering the application of Machine Learning to model an entry table for disability retirement. It is expected that the study’s outcomes will stimulate further scientific research in this field and contribute to the construction of other biometric tables, such as general mortality tables, disabled tables, morbidity tables, and more.
