HealthPredictor

CIS 579 Group 1 Healthcare Project

Setup

• Install git - see here for Windows

• Clone the repo from GitHub: git clone https://github.com/sockduct/HealthPredictor.git

• Install Python 3.13

  • Note: As of February 10, 2025 - the CatBoost Python package used in this Jupyter Notebook is not compatible with Python 3.13. See here for details.

• Install Python 3.12, 64 bit - see here for Windows as of February 10, 2025

• Setup and activate a Python virtual environment:

  • Windows:

    PS> py -3.12 -m venv .venv
    PS> .venv\scripts\activate

• Install required libraries from within Python virtual environment from the root directory of the repository (where the master requirements.txt package specifications is): pip install -r requirements.txt

• Create a .env file with your Azure OpenAI credentials/environment in CKD_Web/Backend:

  OPENAI_API_BASE="https://<YOUR-ENDPOINT.EXAMPLE.COM>/azure-openai-api"
  OPENAI_API_KEY="<YOUR_API_KEY>"
  OPENAI_ORGANIZATION="<YOUR-ORG-ID>"
  MODEL="gpt-4o-mini"
  API_VERSION="2024-06-01"

• Install Node.js (if not installed):

  • Follow this guide to install Node.js.

• Install Node Dependencies in the CKD_Web/UI directory (where the package.json dependency requirements file is):

  PS> npm install

Explore Notebooks

• Start the Jupyter lab environment to browse/work on the Notebooks (from the root directory of the repository): jupyter lab

• Dataset Exploration:

  • Analyze Runs - 100 Random Sampling (Original Dataset)
  • Analyze Runs - 100 KNN Imputation (Original Dataset)
  • HealthPredictor Notebook - Original Dataset
  • HealthPredictor Notebook 2 - New Dataset

• Note: Analyze Runs data created using batch_collector.py

Try Out the Web App in Development Mode (CKD_Web)

Kidney Disease Prediction Backend API (CKD_Web/Backend)

This is a Flask-based REST API that uses a trained Random Forest model to predict the likelihood of chronic kidney disease based on medical input data.

Features

• Accepts JSON data through a POST request
• Predicts using a trained Random Forest model ('CKD-XGB.pkl')
• Returns the prediction serialized in JSON format
• CORS enabled for cross-origin requests

How to Run the Backend

• Make sure you have the trained model file CKD-XGB.pkl in the CKD_Web/Backend directory

• Validate .env correctly configured in the CKD_Web/Backend directory with Azure OpenAI LLM model details and credentials (see Setup section above)

• Run the Flask app using the development server - app runs on http://localhost:5050:

  # Ensure you are running from within the activated Python virtual environment (see Setup)
  PS> python .\app.py

API Endpoint

POST /predict

Kidney Disease Prediction Frontend UI (CKD_Web/UI - Next.js)

This is a frontend application built with Next.js that provides a user interface for the Kidney Disease Prediction API.

How to Run the Frontend

• Ensure Node.js and this app's dependencies are installed (see Setup)

• Start the development server - app runs on http://localhost:3000:

  PS> npm run dev

Frontend App Features

• User form to input medical data
• Sends data to Flask API Backend (/predict endpoint)

Test Web App

• Open a new tab running the webapp:

  • Open Local Web App in New Tab
  • Note: The first time you open the web app and the first time you run each operation, it can take up to a minute for the app to "compile" and run...

• Patient with CKD (Patient ID - 249 from data/chronickidneydisease-uci-400x25.csv)

  • specific_gravity: 1.01
  • albumin: 4
  • hemoglobin: 3.1
  • hypertension: 1
  • serum_creatinine: 13.3
  • sodium: 124
  • white_blood_cell_count: 5400
  • blood_glucose_random: 176
  • packed_cell_volume: 9
  • age: 56

• Patient without CKD (Patient ID - 250 from data/chronickidneydisease-uci-400x25.csv)

  • specific_gravity: 1.025
  • albumin: 0
  • hemoglobin: 15
  • hypertension: 0
  • serum_creatinine: 1.2
  • sodium: 135
  • white_blood_cell_count: 10400
  • blood_glucose_random: 140
  • packed_cell_volume: 48
  • age: 40

• Example chatbot prompts for CKD patients:

  • Why do I have chronic kidney disease?
  • What is the biggest contributor?
  • What are early signs of kidney problems?
  • What foods should I limit with CKD?

• Example chatbot prompts for non-CKD patients:

  • How can I keep my kidneys healthy?
  • Where can I learn more about CKD?

• Example chatbot prompts the will be denied:

  • Where can I get the best deal on Detroit Red Wings tickets? (Non CKD question)
  • Why do I have diabetes? (Non CKD question)

Troubleshooting

• Ensure both the Flask backend and Next.js frontend are running for the app to work correctly
• Ensure you start each one in the correct directory (see notes above)
• Ensure you have followed the setup directions to correctly install Python and Node/JavaScript dependencies
• Ensure you are running the Python backend from the correct activated virtual environment
• The first time you perform each operation on the web site - the first visit, the first patient data submission, etc. - it can take up to a minute each for the app to "compile." If you view the console where you did the npm run dev to run the web app, you'll see it's progress...