Random Forest Classifier for Respiratory Mortality Analytics
Respiratory diseases remain a major contributor to hospital morbidity and mortality worldwide, particularly among elderly patients and individuals with severe pulmonary compromise. Accurate prediction of respiratory mortality is clinically important for triage, resource allocation, ICU utilization, and early intervention. Traditional statistical models frequently demonstrate limited predictive sensitivity because respiratory mortality is influenced by complex interactions among demographic, diagnostic, physiologic, and severity-related variables. In this study, a machine learning framework was developed to predict in-hospital mortality among patients with respiratory disease using administrative and clinically derived variables, including age, sex, length of stay (LOS), diagnostic descriptions, risk of mortality and severity scores. A Random Forest classifier with balanced class weighting was developed and implemented to address nonlinear relationships and class imbalance within the dataset. Initial modeling demonstrated good overall discrimination performance, with receiver operating characteristic area under the curve (ROC-AUC) values approaching 0.84; however, mortality recall remained limited because deceased patients represented a minority class within the original dataset. To improve mortality detection, a physiologically informed synthetic augmentation strategy was developed. Synthetic clinical variables included oxygen saturation, ICU status, ventilator support, sepsis status, systolic blood pressure, creatinine, and lactate levels. Conditional physiologic consistency rules were incorporated during augmentation to preserve clinically plausible relationships among respiratory failure, hemodynamic instability, and organ dysfunction. The augmented dataset substantially improved model sensitivity and balanced mortality classification performance. Final model evaluation demonstrated strong predictive capability, achieving approximately 97% classification accuracy with balanced precision and recall across mortality classes. Confusion matrix analysis revealed marked reduction in false-negative mortality predictions compared with baseline modeling approaches. Feature importance analysis identified physiologic instability markers, respiratory severity classifications, LOS, and diagnostic respiratory categories as dominant predictors of mortality. These findings suggest that hybrid simulation-augmented machine learning frameworks may provide a valuable strategy for respiratory mortality analytics, particularly in datasets with limited real-world mortality prevalence and incomplete physiologic representation.