GJO-XGBoost: Optimized XGBoost with Golden Jackal Algorithm for 30-Days Mortality in Sepsis Patients

Sepsis is a critical condition that remains a global health concern due to its high mortality rate and unpredictable progression.Accurate prediction of 30-day mortality in sepsis patients is pivotal for timely clinical intervention and effective resource allocation.This study introduces an innovative framework that integrates the Golden Jackal Optimization (GJO) algorithm with the XGBoost model to enhance the prediction accuracy of sepsis mortality, utilizing the MIMIC-III dataset.The GJO algorithm was applied to optimize the hyperparameters of the XGBoost model, effectively exploring complex search spaces in high-dimensional data.The model's performance was validated through 5-fold cross-validation and compared against traditional models and advanced ML models such as LightGBM and CatBoost.The GJO-XGBoost model achieved an AUC-ROC of 0.993 and an accuracy of 98.95%, outperforming the baseline XGBoost model (94.71%) and other machine learning models like LightGBM and CatBoost.With precision and sensitivity rates of 99.12% and 98.87%, respectively, the model demonstrated high reliability and generalizability.External validation using the eICU Collaborative Research Database demonstrated the model's robustness and generalizability, achieving an accuracy of 98.51%.Feature importance analysis using SHAP values provided insights into key predictors, ensuring clinical relevance and interpretability.The GJO-XGBoost framework shows promise as a clinical decision-support tool for enabling timely interventions in high-risk sepsis cases.