SEP-XTree: An Explainable AI Model for Early Sepsis Detection
A. J., Madhunisha Mohan, Saravanan Parthasarathy, Vaishnavi Jayaraman, H. Faheem Nikhat · 2025
Sepsis is a life-threatening condition that frequently results in tissue damage, organ failure, and mortality as a result of the body's extreme response to infection. It is imperative to implement early detection to enhance patient outcomes and alleviate healthcare burdens. A groundbreaking model called SEP-XTree was developed for early sepsis prediction. Data preprocessing, mean imputation for missing values, correlation-based approaches for handling missing values, and Recursive Feature Elimination (RFE) for key feature selection are all components of the proposed methodology. To make sure there was a fair representation of both sepsis and non-sepsis patients, the SMOTE (Synthetic Minority Oversampling Technique) was used to rectify class imbalance. After analyzing six different machine learning models, the Extra Trees Classifier stood out with an exceptional 99% F1-score, 99.5% recall, and accuracy. The robustness and absence of overfitting in the model were verified by cross-validation. Along with shedding light on the significance of features, the LIME method improved the model's interpretability. The proposed model exhibits exceptional clinical relevance, accuracy, and reliability for the early detection of sepsis.