EARLY SEPSIS DETECTION USING ENSEMBLE MACHINE LEARNING MODELS

Abstract

Sepsis is a critical medical condition characterized by the body's extreme response to infection, which can lead to organ failure and death if not detected at an early stage. Timely identification of sepsis in intensive care unit (ICU) patients remains a significant challenge due to the complexity and variability of clinical symptoms. This project presents a machine learning-based approach for early prediction of sepsis using ensemble learning techniques such as Random Forest and Bagging Classifier. The system utilizes patient clinical data, including vital signs and laboratory measurements, to build predictive models. The implementation involves essential preprocessing steps such as handling missing values and feature scaling to improve model performance. The trained models are evaluated using standard performance metrics including accuracy, precision, recall, and F1-score, with a focus on reducing false negatives to enhance patient safety. The final model is deployed using a Streamlit-based web application, allowing users to input patient data and obtain real-time predictions. Additionally, the project incorporates basic MLOps practices, including model saving, configuration management, and structured workflow, ensuring scalability and reproducibility. The developed system aims to assist healthcare professionals in early detection of sepsis, thereby improving decision-making, reducing mortality rates, and enhancing overall patient outcomes.