Modelo de Machine Learning para previsão de alta no Hospital Unimed Natal

Abstract

This work addresses the development of a supervised machine learning model using classification techniques to predict the duration of a patient's stay at Hospital Unimed Natal (HUN), based on admission data generated in the company's management system (Philips Tasy). The model, based on the Random Forest algorithm and optimized through Bayesian methods, achieved a Matthews Correlation Coefficient (MCC) of 0.5921 and an accuracy of 0.7413 when evaluated on the test dataset. Despite some limitations, such as constraints on the dataset size and class imbalance, the model demonstrates promising results in classifying patient lengths of stay, offering potential applications in discharge planning and resource management. This research emphasizes the potential of machine learning to enhance hospital operations, optimizing resource allocation and improving service efficiency.