<![CDATA[Hospital length of stay (LoS) prediction is critical for operational planning and resource allocation in stroke care. This study developed and validated machine learning models for predicting binary LoS (short <=7 days vs prolonged >7 days) at a Type A teaching hospital in Indonesia. Random Forest and Gradient Boosting algorithms were trained on 250 stroke patients admitted in 2023 and validated on an independent test set of 62 patients, using 42 clinical features. Gradient Boosting achieved 77.4% accuracy, 78.6% sensitivity, 77.8% specificity, and AUC-ROC of 0.849, outperforming Random Forest by 4.8 percentage points (p=0.008). Hyperparameter optimization via grid search improved baseline accuracy from 74.2% to 77.4%. Feature importance analysis identified primary stroke diagnosis (18.7%), serum AST (12.4%), consciousness level (9.8%), patient age (8.6%), and ICU admission (7.3%) as dominant predictors. Cross-validation showed minimal overfitting, confirming strong generalizability. The optimized model demonstrated high computational efficiency (8.6-second training, 1.9-millisecond prediction per patient), suitable for real-time clinical deployment. This study contributes a validated, locally-optimized tool for Indonesian healthcare, supporting operational decisions on bed allocation, rehabilitation planning, and staff scheduling, with broader applicability to similar lower-middle-income healthcare systems.]]>
Copyrights © 2026
