Predicting patient discharge readiness using machine learning on clinical and operational data
American International Journal of Sciences and Engineering Research · 18.06.2026 · 9 (1) · 1-11
DOI: 10.46545/fze96p32 ↗Kurz gefasst
Fachartikel in der Zeitschrift American International Journal of Sciences and Engineering Research, online erschienen am 18.06.2026. Verfasst von Md Tawfiqul Islam und 2 weiteren Autor:innen. Die Arbeit nennt 12 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Inappropriate prediction of hospital length of stay (LOS) and patient discharge readiness are essential concerns for management as it can affect bed availability, flow in the emergency department (ED), staffing, treatment planning and discharge coordination. Making a prediction of LOS is difficult due to the interrelated nature of several clinical and operational factors affecting hospital stay: age, seriousness of illness, risk of death, admission type, diagnosis group, payment type and status on emergency admission. In this study, a machine learning framework to predict hospital LOS and readiness to be discharged by large scale inpatient data is developed and evaluated. This study employs a dataset of inpatient discharges from the New York State Department of Health for 2,101,588 discharges and 33 variables. To enhance the quality of the data 63,824 deceased-patient records and irrelevant fields were deleted prior to model training. The methodological process consists of data cleaning, missing value treatment, filtering of invalid records, ordinal encoding, feature engineering, normalization, handling class imbalance, splitting the data into a training set and a test set, hyperparameter tuning, and performance evaluation. Several algorithms were compared: Decision Tree, Random Forest, Support Vector Machine, AdaBoost, Gradient Boosting, Logistic Regression, K-Nearest Neighbors, Gaussian Naïve Bayes, XGBoost and CNN. The accuracy of the highest models is found to be 0.93, these are Random Forest, SVM, Gradient Boosting, Logistic Regression, KNN, CNN models. Random Forest achieved a precision of 0.93, recall of 1.00, and F1 score of 0.97, showing consistent predictive results. Based on accuracy, Decision Tree out performed AdaBoost with 0.92 and Gaussian Naïve Bayes with 0.84. The result indicates that ensemble-based models have been effective but Random Forest (RF) is the most effective model for the prediction of LOS, whereas criticality of illness, mortality risk, age group, emergency indicator, diagnostic classification and variables related to admission are the main predictors. JEL Classification Codes: I18, C45, C53.
Zitieren
APA
Islam, M. T., Sarker, M. R., & Rahman, I. (2026). Predicting patient discharge readiness using machine learning on clinical and operational data. American International Journal of Sciences and Engineering Research, 9(1), 1-11. https://doi.org/10.46545/fze96p32
Vancouver
Islam MT, Sarker MR, Rahman I. Predicting patient discharge readiness using machine learning on clinical and operational data. American International Journal of Sciences and Engineering Research. 2026;9(1):1-11. doi:10.46545/fze96p32
BibTeX
@article{Islam2026,
author = {Islam, Md Tawfiqul and Sarker, Md. Rasel and Rahman, Iffat},
title = {Predicting patient discharge readiness using machine learning on clinical and operational data},
journal = {American International Journal of Sciences and Engineering Research},
year = {2026},
volume = {9},
number = {1},
pages = {1--11},
doi = {10.46545/fze96p32}
}
EUVIMED