Hospital Length of Stay and Discharge Prediction Based on Clinical and Demographical Variables: IFMBE Scientific Challenge

Zavaleta-Cavero J.A.; Flores-Perez M.A.; Cieza Huané L.Y. · IFMBE Proceedings (2026) · DOI: 10.1007/978-3-032-20291-8_42

Resumen

Early prediction of hospital outcomes can enhance clinical decision-making and resource planning. This study presents a machine learning framework developed for the IFMBE Scientific Challenge 2025 to predict patient survival and hospital length of stay (LOS) using admission data from over 13,000 patients. Four models were evaluated: K-Nearest Neighbors (KNN), Random Forest (RF), XGBoost (XGB), and a multitask Deep Neural Network (DNN). The dataset included clinical and admission variables collected at hospital entry. Data preprocessing addressed heterogeneity and missing values, while feature selection used non-parametric statistical tests. Models were assessed with F1-score for classification and mean absolute error (MAE) for regression, combined into a global score (GScore). The multitask DNN outperformed other models, achieving the highest F1-score (0.877) and lowest MAE (9.719), resulting in the best GScore (10.949). Despite good performance in survival prediction, LOS estimation remained difficult due to its complex, multifactorial nature. These findings support the use of AI-based tools, particularly multitask learning, for early outcome prediction in hospital settings. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.

Instituciones: Universidad Peruana Cayetano Heredia; Pontificia Universidad Católica del Perú