ManvithMogaveera/Sepsis-Early-Detection-using-Machine-Learning-on-Real-ICU-Patient-Data

# 🚑 SepsisSense: Early ICU Sepsis Prediction using Temporal Machine Learning      # 📌 Problem Statement Sepsis is one of the leading causes of mortality in Intensive Care Units (ICUs). Delayed detection can rapidly result in: * Organ failure * Septic shock * Multi-organ dysfunction * Increased ICU mortality Traditional diagnosis often occurs after significant physiological deterioration. This project focuses on leveraging Machine Learning and temporal clinical data to assist in identifying early warning signs of sepsis before critical deterioration occurs. #🔗WEBSITE https://sepsissense.streamlit.app/ Please,do checkout! # 🧠 Project Overview SepsisSense was developed using a large-scale real-world ICU dataset containing over **1.5 million patient records** with temporal physiological measurements such as: * Heart Rate (HR) * Blood Pressure (MAP/SBP/DBP) * Respiratory Rate * Oxygen Saturation * Laboratory Biomarkers * ICU Progression Data The system performs: * Leakage-safe patient-wise preprocessing * Temporal feature engineering * Missing value handling for sparse ICU data * Imbalance-aware classification using XGBoost * Real-time risk prediction through Streamlit deployment # ⚙️ Key Features ## ✅ Temporal Healthcare Preprocessing * Patient-wise interpolation * Forward filling of clinical measurements * Leakage-safe GroupShuffleSplit validation * Sparse feature handling for ICU lab tests ## ✅ Clinical Feature Engineering Engineered medically meaningful features such as: * Shock Index * Pulse Pressure * Rolling Mean Statistics * Physiological Trend Differencing * HR × Temperature Interaction ## ✅ Imbalanced Learning Strategy * Handled severe class imbalance using: * `scale_pos_weight` * Recall-oriented evaluation * Threshold optimization ## ✅ Real-Time Prediction Ready * Streamlit-powered interactive dashboard * Live sepsis risk probability prediction * Clinical decision support style output # 📊 Model Performance | Metric | Score | | ---------------- | ---------------------------- | | ROC-AUC | **0.825** | | Sepsis Recall | **68%** | | Optimized Recall | **90%** | | Dataset Size | **~1.5 Million ICU Records** | ### 📌 Clinical Interpretation The model prioritizes **high recall** to reduce the probability of missing septic patients, which is critical in real-world healthcare scenarios where delayed intervention can be life-threatening. # 🧪 Tech Stack Python Pandas NumPy Scikit-learn XGBoost Matplotlib Streamlit # 📁 Project Structure SepsisSense/ │ ├── app.py ├── preprocessing.py ├── model_1.py ├── EDA_analysis.ipynb ├── requirements.txt ├── README.md ├── LICENSE └── images/ # 📷 Project Visualizations ## 🖥 Streamlit Dashboard ## 📈 ROC Curve (Add ROC curve image here) ## 📊 Feature Importance (Add feature importance graph here) ## 🧾 Confusion Matrix (Add confusion matrix image here) # 🚀 Installation git clone https://github.com/ManvithMogaveera/Sepsis-Early-Detection-using-Machine-Learning-on-Real-ICU-Patient-Data.git cd SepsisSense pip install -r requirements.txt # ▶️ Run the Application streamlit run app.py # 🔬 Dataset Dataset used: **PhysioNet / Computing in Cardiology Sepsis Challenge Dataset** Due to dataset size and licensing constraints, the original dataset is not uploaded to this repository. # 🌍 Real-World Impact This project demonstrates how Machine Learning can assist healthcare professionals in: * Early critical care risk detection * Continuous ICU patient monitoring * Clinical decision support systems * Reducing delayed sepsis diagnosis The pipeline was intentionally designed with healthcare-safe preprocessing and leakage prevention techniques to better simulate real-world ICU deployment scenarios. # 📌 Future Improvements * SHAP Explainability Integration * Real-Time ICU Monitoring Support * Transformer/LSTM Temporal Models * Cloud Deployment * Live Clinical Alert System # 👨‍💻 Author ### Manvith Mogaveera AI & Machine Learning Enthusiast focused on: * Healthcare AI * Clinical ML Systems * Real-World Intelligent Applications * Temporal Machine Learning ⭐ If you found this project interesting, consider starring the repository.