Analysis On Mortality Rate Due to Sepsis in Different Demography

Abstract The incidence of sepsis, which accounts for up to 6% of all hospital admissions, is estimated to be between 250 and 500 cases per 10,000 people per year. Sepsis is the condition that costs the healthcare system the most and has been identified as one of the most pressing issues affecting global health. So, our study provides a solution to analyze the chances of the occurrence of sepsis by measuring the SIRS Score of the patients using their physiological characteristics of the patients, such as heart rate, respiratory rate, white blood cell count and body temperature. Our model is able to predict the outcome as dead or alive according to certain characteristics so that healthcare specialists can prioritize treatment for patients accordingly. Due to the highly imbalanced nature of the dataset, several techniques such as Random Oversampling and Random Undersampling were applied to address class imbalance. A Logistic Regression classifier was trained, and performance was evaluated using various metrics, including accuracy , precision, recall, F1-score, and the confusion matrix. Initially, the model showed high precision but poor recall for the sepsis class, which was improved by adjusting the decision threshold from 0.5 to 0.4. Cross-validation and AUC score evaluations demonstrated a solid model performance, with AUC consistently above 0.7. To further enhance the model, hyperparameter tuning and alternative models like Random Forest and XGBoost were considered. The findings highlight the importance of threshold adjustment, cross-validation, and feature 1 selection for improving model performance, particularly in imbalanced classification tasks. Further improvements could be made by fine-tuning hyperparameters and exploring additional machine learning models.