Using machine learning to evaluate the application of resources in hospital inpatient care

Abstract

The article addresses the use of machine learning to optimize resource allocation in Hospital Admission Assistance within Brazil’s Unified Health System (SUS), considering the management challenges posed by decentralization and demographic transitions. The main objective is To present a machine learning method to support managers in evaluating discrepancies between predictions and resource allocation in Hospital Admission Assistance. Data from Datasus (2022-2024) were preprocessed and used to train regression models, including Decision Tree, Random Forest, and MLP (Multi-Layer Perceptron), focusing on predicting average hospitalization costs per municipality. The MLP model stood out with the lowest errors (MAPE: 28.46% and MAE: 183.20). Results indicate that the model can support strategic decisions, enabling local and regional analyses for greater efficiency in public spending. It is concluded that the approach contributes to prioritizing resource allocations and suggests improvements in public policies based on artificial intelligence.