Mortality Prediction of Sepsis using MIMIC-III Dataset

Abstract

Sepsis is a potentially fatal illness marked by an unbalanced host response to infection, which results in organ failure. Early detection of individuals at mortality risk is essential for prompt management and better patient outcomes. To effectively forecast mortality in septic patients, this study will create and contrast prediction models using the XGBoost and Random Forest algorithms. The clinical characteristics of sepsis patients, including their demographic data, vital signs, test findings, and comorbidities, are included in the dataset used in this study. To guarantee data quality and consistency, data preprocessing techniques like the imputation of missing values and normalization were used. The created models' predictive skills were evaluated using performance assessment criteria like accuracy, precision, recall, and area under the receiver operating characteristic curve (AUC-ROC). According to the findings, the XGBoost and Random Forest algorithms both performed effectively in forecasting septic patient death. High accuracy, precision, recall, and AUC-ROC scores of the models suggested their potential as trustworthy instruments for determining mortality risk in clinical practice.