BioMetric AI - Advanced Obesity Risk Assessment

BioMetric AI uses Stacking Ensemble ML to predict obesity risk with 77% accuracy. By analyzing lifestyle and clinical data, it delivers instant health insights, enabling proactive disease prevention.

Comment

Inspiration

The World Health Organization reports that 1 in 8 people live with obesity, a major risk factor for chronic illness. BioMetric AI was inspired by the urgent need to shift healthcare from 'treatment' to 'prevention.' We leveraged advanced Machine Learning to create a tool that makes professional-grade risk assessment free, instant, and accessible to anyone with a browser, removing the barriers between people and their health data.

What it does

  • Collects lifestyle and physiological inputs via a simple form (age, height, weight, diet, activity, comorbidities, etc.).
  • Validates and preprocesses inputs using the same deterministic pipeline used during training.
  • Runs a Stacking Ensemble model to classify a user's obesity risk and returns risk category.

How we built it

  • Data & model: feature engineering and model development were performed using scikit-learn utilities and CatBoost base learners; final model is a stacking ensemble was saved.
  • Backend: FastAPI exposes inference endpoints, performs input validation (Pydantic), and loads preprocessing artifacts at startup for fast prediction.
  • Frontend: React + Vite provides a responsive UI, form validation, and visualization of prediction scores.
  • Dev workflow: artifacts and column-order metadata are versioned; training notebooks and preprocessing scripts are kept for reproducibility.

Challenges we ran into

  • Class imbalance and real-world skew — we addressed this using stratified CV, class-weighting and careful calibration.
  • Feature drift and schema mismatch — keeping model_columns.joblib and encoders synchronized between training and serving required strict versioning.
  • Trade-offs between model complexity and latency — stacking improved accuracy but required careful optimization to keep inference times low.
  • Privacy & data minimization — designing the API to avoid storing PII while still providing useful, actionable feedback.

Accomplishments that we're proud of

  • A reproducible ML pipeline that produces a robust stacking ensemble with strong cross-validated performance.
  • Fast, user-friendly UI that gives immediate, feedback to users.
  • Production-oriented API.

What we learned

  • Small, careful preprocessing decisions materially affect model stability in production.
  • Explainability and communication matter — even accurate predictions need clear guidance to be useful to non-experts.
  • Automating artifact versioning (model columns, encoders, scaler) prevents many serving-time issues.

What's next for BioMetric AI - Advanced Obesity Risk Assessment

  • Monitoring and model drift detection to automatically flag when retraining is necessary.
  • Add explainability (SHAP or similar) to surface feature contributions and increase clinician trust.
  • Clinical validation studies and partnerships to evaluate real-world utility and safety.
  • Improve accuracy with more data.

Built With

Share this project:

Updates