WCSE

A Comparative Study of Machine Learning Algorithms for 30-Day ICU Readmission in Heart

Abstract— Early readmission of heart failure (HF) patients within 30 days following intensive care unit (ICU) discharge remains a significant clinical and economic challenge. Accurate prediction of this risk is crucial for improving patient management and optimizing resource utilization. This retrospective study analyzed data from 5,414 adult HF patients admitted to the ICU within the MIMIC-III database. To address class imbalance, advanced oversampling techniques were applied during training. Eight machine learning algorithms were implemented and evaluated for 30-day ICU readmission prediction. Model performance was assessed using accuracy and AUROC. Among the eight algorithms, LightGBM achieved the highest individual performance with an accuracy of 88.72% and an AUROC of 74.48% for predicting 30-day ICU readmission. The ensemble model, leveraging top-performing algorithms including LightGBM, demonstrated enhanced predictive capabilities. Key variables, such as vital signs and comorbidities, were identified as critical predictors. LightGBM demonstrates strong potential for accurately predicting 30-day ICU readmission in HF patients. The high performance of LightGBM and the benefits of ensemble methods suggest these tools can effectively identify high-risk patients, enabling targeted interventions to reduce readmissions and improve care. Future research should focus on external validation and clinical implementation.

Index Terms— heart failure; intensive care units; machine learning; readmission prediction; electronic health records, readmission.

Chuan-Mei Chu, Hong-Yan Chen, Te-Nien Chien
College of Management, National Taipei University of Technology, Taiwan
Chen-Shug Wang
Department of Information and Finance Management, National Taipei University of Technology, Taiwan
Bo-Yi Li
Department of Management Information Systems, National Chengchi University, Taiwan

[Download]

Cite: Chuan-Mei Chu, Chen-Shug Wang, Hong-Yan Chen, Bo-Yi Li, Te-Nien Chien, "A Comparative Study of Machine Learning Algorithms for 30-Day ICU Readmission in Heart", 2025 the 15th International Workshop on Computer Science and Engineering (WCSE 2025), pp. 245-249, Jeju Island, South Korea, June 28-30, 2025.