Smart budgeting in the SUS (Brazilian Unified Health System)

A new artificial intelligence-based model for classification and forecasting of hospital expenses

Abstract

This study addresses the challenges of budget management within Brazil’s Unified Health System (SUS), particularly regarding the forecasting and clas- sifying of municipal expenditures on hospital admissions, which affect the ef- ficiency and equity of public health financing. It proposes an intelligent bud- geting model based on machine learning, using data from Datasus (2022–2024) to train algorithms such as Naïve Bayes, Random Forest, and Multi-Layer Per- ceptron (MLP). The results show that Naïve Bayes achieved superior perfor- mance in expenditure classification, with a Kappa index of 0.933 and an area under the ROC curve of 0.992, while the MLP demonstrated greater accuracy in hospital cost forecasting, significantly reducing absolute and percentage er- rors. It is concluded that the use of predictive and classificatory models based on artificial intelligence optimizes resource allocation, promoting transparen- cy, efficiency, and sustainability in public health financing, while reinforcing the strategic role of the State in ensuring universal and equitable services. The main contribution of this work lies in the proposal of an innovative intelligent budgeting system, which challenges neoliberal narratives advocating for the reduction of the State by demonstrating how advanced technologies can stren- gthen public administration.