AI-Driven Forecasting of Patient Discharge Timelines with Transparent Insights

Abstract

Effective bed management in hospitals reduces expenses and enhances patient outcomes. This research introduces a predictive model for ICU length of stay (LOS) at the time of admission, utilizing electronic health records (EHR) information. The examine assesses multiple machine learning methods, along with Logistic Regression, Random forest, MLP, Gradient Boosting, XGBoost, and an extension with CatBoost, utilizing the clinic stay dataset from the Kaggle repository. The algorithms are evaluated the usage of AUC, accuracy, precision, consider, and F1-score. XGBoost attained the best accuracy among conventional algorithms, but the stronger CatBoost algorithm passed all others with an accuracy of 98.25%. methods of Explainable AI (XAI), together with SHAP, were hired to elucidate characteristic contributions. The research illustrates the capability of using patient EHR statistics and complicated machine learning models to forecast ICU admissions, facilitating progressed useful resource distribution in healthcare centers.