IAAHS JoCo 2025 – Episode 7: E-Mergency Room: Predictive Modeling of Supplementary Healthcare Expenses using Machine Learning and Deep Learning Techniques

With the outbreak of Covid-19 pandemic, the Brazilian supplementary healthcare sector became a conducive environment for using complex data analysis and modeling tools. In this study, we apply different Machine and Deep Learning techniques (SVM, XGBoost and RNN) to predict healthcare expenses and evaluate if these techniques would present better performance in comparison to traditional ones, such as time series and regressions. Prediction scenarios were generated upon expense official databases between 2015-2022, considering two panoramas: (i) real, and; (ii) counterfactual, in which we assume the non-existence of the pandemic data for 2020. Using RMSE as the performance indicator, we find out that XGBoost model presented the best performance for the real panorama, with better fit in 32.2% of the scenarios. For the counterfactual panorama, we observe that RNN and SVM models obtained better fit in 22.3% of the cases. It is noteworthy that, until now, no studies were identified that address the use of predictive Machine and Deep Learning models into the Brazilian healthcare expenses. We also expect that this study offers insights for decisions made by the several players in this sector, such as operators and regulators, especially when it comes to pricing and development of healthcare products.