AyurHealth – Disease Prediction and Remedies (Flask)

1) What this project is

AyurHealth is a small Flask web app that:

• Predicts a probable disease from 4 user-entered symptoms using ML
• Suggests Ayurvedic remedies for a selected condition from a curated CSV
• Provides a simple, responsive UI (Bootstrap + custom CSS)

2) How it works (high level)

• The Symptoms page posts four symptoms to the backend.
• The backend trains multiple classifiers on Training.csv (RandomForest, SVC, LogisticRegression, GaussianNB), evaluates them on Testing.csv and selects the model with the best accuracy for prediction.
• The predicted disease is displayed and also stored in a local SQLite DB.
• The Suggester page looks up Remedies.csv and displays the matching row.

3) Run locally

Prerequisites: Python 3.10+ recommended

a. Create/activate a virtual environment (optional but recommended)

python -m venv .venv
.venv\Scripts\activate      # Windows
source .venv/bin/activate    # macOS/Linux

b. Install dependencies

pip install -r requirements.txt

c. Start the app

python AyurHealth.py

d. Open in browser

Navigate to: http://localhost:3000

4) App navigation

• Home → Overview and quick links
• Symptoms → Enter four symptoms (pretty labels), get ML prediction
• Suggester → Select a condition and view remedies from CSV
• About → About page

5) Important files and folders

Core Application Files

• AyurHealth.py

  • Flask application entry point and routes:
    • / → Home
    • /symptoms → Symptoms page (GET)
    • /suggester → Suggester page (GET)
    • /about → About page (GET)
    • /s1 (POST) → Handles symptoms form, predicts disease and renders result
    • /s2 (POST) → Handles suggester form and renders remedies
  • The server runs on host 'localhost', port 3000 (debug=True)

• Symptom.py

  • ML workflow for disease prediction:
    • Loads Training.csv / Testing.csv
    • Encodes prognosis labels to integers
    • Trains candidate models: RandomForest, SVC, LogisticRegression, GaussianNB
    • Evaluates on Testing.csv, selects the best by accuracy
    • Builds a feature vector from the four symptoms and predicts with the best model
    • Logs model accuracies and chosen model
    • Saves each prediction into SQLite (database.db → table SymptomPrediction)

• Suggester.py

  • Simple CSV lookup for remedies. Finds the row in Remedies.csv whose first cell matches the chosen condition and returns the formatted content.

Templates (Jinja2)

• templates/index.html → Home (navbar includes Symptoms and Suggester)
• templates/symptoms.html → Symptoms form + prediction result
• templates/suggester.html → Remedies lookup + result
• templates/about.html → Static info page

Static Assets

• static/css/
  • project.css → Custom styles and responsive rules
  • bootstrap.min.css → Bootstrap framework
• static/js/
  • slide.js → Home page image slideshow (only runs on Home)
  • symptoms.js → Hides previous prediction when user edits inputs
  • suggester.js → Page-specific JS for suggester (if any hooks)
• static/images/ → App images and icons

Data Files

• Training.csv → Training data with symptom columns and a prognosis column
• Testing.csv → Test set used for accuracy selection
• Remedies.csv → Conditions and corresponding remedy text

Configuration & Database

• database.db

  • Local SQLite DB created at runtime; stores symptom inputs and predictions
  • Ignored by Git via .gitignore

• requirements.txt

  • Minimal dependencies to run the app (Flask, pandas, scikit-learn)
  • If your platform needs it, install numpy/scipy as well (pip will usually resolve)

• start.sh

  • Simple one-line runner used for Glitch/hosting scenarios

• .gitignore

  • Excludes caches, virtualenvs, databases, build outputs and editor files

6) UI details / helpful behaviors

• Symptoms inputs show human-friendly labels (e.g., “Back Pain”) but the backend receives machine codes (e.g., back_pain) so the model works consistently.
• After submitting Symptoms/Suggester, the page anchors to the result box.
• When the user focuses/types again in Symptoms inputs, the previous result box is hidden to avoid confusion.

7) Notes on data and accuracy

• Accuracy is selected on a small Testing.csv (43 rows), so different models can tie. The project logs per-model accuracies and the selected model each request.
• For stronger validation, consider k-fold cross-validation on Training.csv and using Testing.csv strictly for a final holdout check.