Welcome to the Glucose Level Prediction project. This repository provides a machine learning solution to predict future blood glucose levels based on physiological signals using XGBoost. It also includes a Streamlit web application that allows users to interact with the model in real-time.

1. Overview
2. Dataset Description
3. Data Preprocessing
4. Model Architecture
5. Evaluation Metrics
6. Sample Results
7. Web Application (Streamlit)
8. Getting Started
9. Folder Structure
10. License

This project aims to predict the blood glucose level (mmol/L) one hour into the future using a variety of physiological time-series features such as:

• Insulin intake
• Carbohydrate consumption
• Heart rate
• Step count
• Calories burned
• Activity level

The goal is to assist individuals and healthcare professionals in proactively managing blood glucose levels.

The dataset consists of time-lagged features representing past physiological readings. For instance:

• bg-0:15, hr-0:30, insulin-1:00, etc. represent the value at 15 minutes, 30 minutes, or 1 hour ago.
• The target column is bg+1:00, representing the glucose level 1 hour into the future.

Preprocessing steps included:

• Feature Aggregation: Time-series features were aggregated (mean, std, etc.).
• Missing Value Handling: Mean imputation grouped by participant (p_num).
• Categorical Conversion: The activity feature was converted to the mode value per row.
• Scaling: Numerical features were optionally scaled for model input.

We use XGBoost Regressor, a powerful gradient boosting model suitable for tabular data.

• Input: Cleaned physiological feature set
• Output: Predicted glucose level (bg+1:00)
• Training: Hyperparameters tuned using cross-validation

These metrics help us assess the model's performance and its deviation from true glucose values.

Visualizations provided:

• ✅ Feature Importance Plot (from XGBoost)
• 📈 Actual vs Predicted Glucose Levels
• 📊 Error Distribution Histogram

These results demonstrate the interpretability and reliability of the model.

An interactive frontend built with Streamlit allows users to input their physiological parameters and get a real-time prediction of their blood glucose level.

• Displays predicted glucose level
• Easy-to-use and lightweight

1. Clone the repository:

git clone https://github.com/arjunravi26/Glucose-Monitoring.git
cd Glucose-Monitoring

2. Install dependencies:

pip install -r requirements.txt

3. Run the Streamlit app:

streamlit run main.py

Glucose-Monitoring/
├── data/                   # Dataset files
├── models/                 # Trained model files
├── notebooks/              # Jupyter notebooks for EDA & training
├── app/                    # Streamlit app files
├── main.py                 # Entry point for Streamlit app
├── utils.py                # Helper functions
├── requirements.txt        # Dependencies
└── README.md               # Project overview

This project is licensed under the MIT License. Feel free to use, modify, and share with credit.

For questions, feedback, or collaboration:

• Author: Arjun Ravi
• GitHub Repo: Glucose-Monitoring

"Empowering personal health through predictive intelligence."