GlucoTrack: Predicting Diabetes Likelihood from Lifestyle and Clinical Data

Live app: glucotrack.streamlit.app

Deployment

Run the Streamlit app locally to interact with the champion model (Balanced FFNN) and get diabetes risk predictions.

Prerequisites:

• Python 3.9+
• Dependencies: streamlit, torch, pandas, numpy, scikit-learn, joblib

Quick start:

# Create and activate a virtual environment (optional but recommended)
python -m venv .venv
source .venv/bin/activate  # Linux/macOS

# Install dependencies
pip install --upgrade pip
pip install streamlit torch pandas numpy scikit-learn joblib

# Launch the app
cd yan-cotta
streamlit run app.py

Notes:

• The app loads artifacts from yan-cotta/deployment_artifacts (model weights and preprocessing objects) generated from the Week 2 and Week 3 notebooks.
• If you update preprocessing or retrain the model, re-run the last cells in those notebooks to refresh the files in deployment_artifacts before restarting the app.

Deploy to Streamlit Cloud

• Ensure these files exist at the repository root:
  • requirements.txt (Python dependencies)
  • runtime.txt (Python version; not always honored on Cloud which may default to Python 3.13)
• In Streamlit Cloud, set the entrypoint to yan-cotta/app.py.
• PyTorch note: Cloud currently uses Python 3.13, so torch==2.8.0 is pinned in requirements.txt because it provides wheels for 3.13.
• After deployment, use “Manage app” → Logs to inspect any missing dependency errors.

🧠 Project Overview

-> shared repository with all team members work: https://github.com/SuperDataScience-Community-Projects/SDS-CP035-gluco-track

GlucoTrack is a health-focused classification project that aims to predict whether an individual is diabetic, pre-diabetic, or non-diabetic using self-reported survey and biometric data. The dataset includes responses collected by the CDC on physical activity, BMI, mental health, general wellness, and other lifestyle variables.

The project is built around a real-world public health challenge: how do we identify individuals at risk before it’s too late? You’ll analyze patterns in health behavior and demographic variables to develop a predictive model that supports early interventions.

This project is split into two experience tracks:

• 🟢 Beginner Track – Build traditional classification models using scikit-learn and deploy a simple Streamlit app.
• 🔴 Advanced Track – Design and train a deep learning classifier with PyTorch or TensorFlow, integrate embeddings for categorical data, and deploy via Docker or Hugging Face Spaces.

---

🧪 Dataset Summary

• Source: CDC Behavioral Risk Factor Surveillance System (BRFSS)
• Instances: ~250,000
• Target Variable: Diabetes_binary (0 = No, 1 = Pre-diabetes or Diabetes)
• Features: 20+ features covering BMI, physical activity, mental health days, general health, smoking status, and more
• Data Type: Tabular, mostly binary and ordinal features
• Task Type: Classification

---

🧰 Tools & Libraries

• ML Libraries: scikit-learn, XGBoost, LightGBM (Beginner); PyTorch or TensorFlow (Advanced)
• Visualization: matplotlib, seaborn
• Deployment: Streamlit (Beginner), Docker + Hugging Face Spaces or Flask API (Advanced)
• Experiment Tracking: MLflow

---

📂 Track Structure

🟢 Beginner Track

➡️ Beginner Scope of Works ➡️ Beginner Report Template ➡️ Submit Your Work

🔴 Advanced Track

➡️ Advanced Scope of Works ➡️ Advanced Report Template ➡️ Submit Your Work

---

🚀 Learning Outcomes

By the end of this project, you will:

• Understand health-related feature engineering and data cleaning techniques
• Apply classification models and evaluate their performance with appropriate metrics

• Track experiments and manage reproducibility using MLflow
• Deploy interactive predictive tools for real-world use cases

---

👥 Who Should Join?

This project is ideal for:

• Aspiring machine learning engineers and data scientists interested in healthcare analytics
• Beginners seeking to master the end-to-end ML workflow

• Intermediate learners who want to experiment with neural networks and sequence modeling

---

📝 Acknowledgements

This project uses open public health data provided by the Centers for Disease Control and Prevention (CDC).