Background/Objectives: This study develops machine learning (ML) models to predict hypoxemia severity during emergency triage, particularly in Chemical, Biological, Radiological, Nuclear, and Explosive (CBRNE) scenarios, using physiological data from medical-grade sensors. Methods: Tree-based models (TBMs) such as XGBoost, LightGBM, CatBoost, Random Forests (RFs), Voting Classifier ensembles, and sequential models (LSTM, GRU) were trained on the MIMIC-III and IV datasets. A preprocessing pipeline addressed missing data, class imbalances, and synthetic data flagged with masks. Models were evaluated using a 5-min prediction window with minute-level interpolations for timely interventions. Results: TBMs outperformed sequential models in speed, interpretability, and reliability, making them better suited for real-time decision-making. Feature importance analysis identified six key physiological variables from the enhanced NEWS2+ score and emphasized the value of mask and score features for transparency. Voting Classifier ensembles showed slight metric gains but did not outperform individually optimized models, facing a precision-sensitivity tradeoff and slightly lower F1-scores for key severity levels. Conclusions: TBMs were effective for real-time hypoxemia prediction, while sequential models, though better at temporal handling, were computationally costly. This study highlights ML’s potential to improve triage systems and reduce alarm fatigue, with future plans to incorporate multi-hospital datasets for broader applicability.

This paper presents the development of machine learning (ML) models to predict hypoxemia severity during emergency triage, especially in Chemical, Biological, Radiological, Nuclear, and Explosive (CBRNE) events, using physiological data from medical-grade sensors. Gradient Boosting Models (XGBoost, LightGBM, CatBoost) and sequential models (LSTM, GRU) were trained on physiological and demographic data from the MIMIC-III and IV datasets. A robust preprocessing pipeline addressed missing data, class imbalances, and incorporated synthetic data flagged with masks. Gradient Boosting Models (GBMs) outperformed sequential models in terms of training speed, interpretability, and reliability, making them well-suited for real-time decision-making. While their performance was comparable to that of sequential models, the GBMs used score features from six physiological variables derived from the enhanced National Early Warning Score (NEWS) 2, which we termed NEWS2+. This approach significantly improved prediction accuracy. While sequential models handled temporal data well, their performance gains didn’t justify the higher computational cost. A 5-minute prediction window was chosen for timely intervention, with minute-level interpolations standardizing the data. Feature importance analysis highlighted the significant role of mask and score features in enhancing both transparency and performance. Temporal dependencies proved to be less critical, as Gradient Boosting Models were able to capture key patterns effectively without relying on them. This study highlights ML's potential to improve triage and reduce alarm fatigue. Future work will integrate data from multiple hospitals to enhance model generalizability across clinical settings.