🧠 Fall Risk Predictor

A fullstack AI-powered web application to assess fall risk in patients with Parkinson's Disease. (coming soon)

🖼️ Demo Preview

🚀 Project Overview

Fall Risk Predictor is a clinically-oriented web application designed to predict the likelihood of falls in Parkinson’s patients using a machine learning model based on real-world clinical and biomechanical features. It provides an intuitive interface, interpretable outputs, and smart tooltips showing normal reference ranges for each parameter.

🧩 Tech Stack

• Frontend: React (with Axios) — deployed on Vercel
• Backend: Django REST Framework — deployed on Render
• Model: Random Forest Classifier with feature selection and scaling
• ML Tools: scikit-learn, joblib, pandas, numpy
• Other: HTTPS CORS handling, custom visual feedback, deployment automation

📊 Features Used for Prediction

Feature Name	Description
hr_v	Harmonic Ratio – Vertical
%det_v	% Determinism – Vertical
%det_ml	% Determinism – Medio-Lateral
%det_ap	% Determinism – Antero-Posterior
weigth	Patient's Weight (kg)
age_onset	Age at disease onset
duration_years	Disease duration in years
ledd	Levodopa Equivalent Daily Dose (mg/day)
updrs-ii	UPDRS-II score (Activities of Daily Living)
updrs-iii	UPDRS-III score (Motor Evaluation)

💡 Key Functionalities

• Clean and responsive UI for clinical environments
• Dynamic tooltips for each feature showing quartile-based reference ranges
• Real-time prediction based on ML inference
• Output includes:
  • Prediction (🟢 Non Fallers / 🟠 Fallers)
  • Fall probability (as a percentage)
  • Color-coded warning for high-risk cases (≥ 88%)

🔄 System Architecture

User (React UI) ↓ Axios Django API @ /api/predict/ & /api/ranges/ ↓ Scikit-learn model (Random Forest)

⚙️ How to Run Locally

Clone the repository

git clone https://github.com/DanteTrb/fall-risk-predictor.git cd fall-risk-predictor

Backend setup

cd backend pip install -r requirements.txt python manage.py runserver

Frontend setup

cd ../frontend npm install npm start

📝 Note: For production, the backend is hosted on Render and the frontend on Vercel. Axios paths are automatically routed correctly in production via proxy rules.

🌐 Live Demo

Frontend: fall-risk-predictor-hazel.vercel.app (Backend hosted on Render — may take up to 50 seconds to wake up if idle.)

📦 Deployment Status

Component	Platform	Status
Frontend	Vercel	✅ Live
Backend	Render	✅ Live
Domain	fallrisk.ai	⏳ Configuring DNS

👨‍⚕️ Author

Dante Trabassi Biomedical Engineer | PhD Neuroscience | Sapienza University of Rome Researcher in AI for Movement Disorders, Clinical Gait Analysis, Generative and xAI Models

📄 License

This project is licensed under the MIT License. See LICENSE for full terms.

🤝 Contributing

Pull requests and suggestions are welcome! If you're a clinician, data scientist, or ML researcher interested in neurodegenerative disorders, neuroscience and fall risk, feel free to collaborate.

This project integrates explainable AI and generative AI techniques into biomechanics and gait analysis, with a focus on Parkinson's disease.