REAL-TIME INFERENCE OF SEPSIS SEVERITY USING WEARABLE BIOSENSORS, GENOMIC MARKERS, AND EXPLAINABLE ENSEMBLE MACHINE LEARNING MODELS IN ICU PATIENTS

Abstract

Sepsis is a life-threatening condition characterized by dysregulated immune response and rapid organ dysfunction, demanding early and accurate prediction for effective clinical intervention. This study proposes an explainable ensemble machine learning model that fuses real-time physiological signals from wearable biosensors, genomic expression profiles, and clinical parameters to predict sepsis severity in ICU patients. The model integrates Gradient Boosting, Support Vector Machines, and Deep Neural Networks through soft voting, enabling robust classification across mild, moderate, and severe cases. With a dataset comprising over 500 ICU patients, the ensemble demonstrated superior performance, achieving an F1-score of 0.92 and ROC-AUC of 0.96. SHAP-based interpretability revealed oxygen saturation, respiratory rate, SOFA score, and genomic inflammation markers as the most influential predictors. Nine detailed tables present biosensor-clinical correlations, model confusion matrices, and severity classification summaries, while twelve complex visualizations—including hybrid plots, boxplots, and cluster maps—validate the model’s performance and insights. The simulated real-time deployment scenario confirmed the model’s adaptability in tracking physiological shifts and updating severity risk scores every 15 minutes. Clinician feedback supported the usability and interpretability of the model outputs. This fusion of multimodal data streams into a transparent, adaptive AI framework marks a significant advancement in sepsis monitoring and offers a scalable solution for personalized, data-driven critical care.