Year: 2025
Type: article
Abstract: Early mortality prediction in Intensive Care Units (ICU) using machine learning techniques has become a key tool for supporting clinical decision-making in highly dynamic and critical healthcare environments. In this study, a Multilayer Perceptron (MLP) neural network model was developed to predict mortality in the ICU of the "Arnaldo Milián Castro" University Hospital in Villa Clara, Cuba. Preprocessing techniques were applied to handle inconsistencies in clinical data, along with balancing methods such as BorderlineSMOTE to address class imbalance. The model achieved notable metrics in field tests: sensitivity above 85%, area under the ROC curve (AUC-ROC) of 0.93, and high overall accuracy. Additionally, the LIME method (Local Interpretable Model-Agnostic Explanations) was implemented to enhance the interpretability of predictions. The results demonstrate the system's potential to identify at-risk patients with minimal error margin, providing a valuable support tool for medical staff in intensive care settings.
Language: Spanish
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7220307247
