PredictPulse

A navigation prediction and asset preloading system that forecasts user navigation paths and preloads web assets using a Decision Tree model.

Overview

PredictPulse enhances web performance by predicting user navigation paths and preloading assets (images, CSS, JS) for the next page, based on interaction patterns, device context, and session data. It uses a client-side Decision Tree model trained on user logs via train_model.py, dynamically adapting to userAgent and screenWidth. The system achieves 54.55% accuracy with a 100-log dataset and introduces the Web Operational Intelligence Score (WOIS) to evaluate adaptive intelligence.

Features

• Navigation Prediction & Preloading: Forecasts next pages and preloads assets using a Decision Tree model.
• Dynamic Adaptation: Adjusts predictions based on userAgent and screenWidth.
• User Context Awareness: Incorporates device type, OS, browser, and screen size.
• Open-Source: Deployed at gitgetgotgotten.github.io/predictpulse.

Init

1. Clone: git clone https://github.com/gitgetgotgotten/predictpulse
2. Go: cd predictpulse
3. Install: npm install
4. Generate mock data: node scripts/generate_mockdata.js (outputs to predictpulse_mockdata.json)
5. Train model: pip install pandas scikit-learn numpy ua-parser && python train_model.py (outputs to src/utils/predictNextPage.js, public/encoders.json, public/asset_map.json)

Deploy

1. Build: npm run build
2. Deploy: npm run deploy
3. Visit: Your https://gitgetgotgotten.github.io/predictpulse equivalent

Redeploy

1. git checkout main
2. Remove-Item -Recurse -Force dist (Windows PowerShell)
   • macOS/Linux: rm -rf dist
3. git push origin :gh-pages
4. python train_model.py
5. python generate_asset_map.py
6. npm run build
7. node scripts/build.js
8. cp public/asset_map.json dist/asset_map.json
9. npm run deploy

Evaluate (Mock)

1. Run: python evaluate.py
2. Check: evaluation_results.json

Evaluate (Real)

1. Run: python evaluate_real.py
2. Check: evaluation_results_real.json

Real-World Data Collection

Logs are stored in IndexedDB and uploaded to a private GitHub repo (predictpulse-data, data branch) after each page visit. To set up:

• Create a private repo: predictpulse-data.
• Create a data branch.
• Generate a GitHub Personal Access Token with repo scope.
• Set token: localStorage.setItem('github_token', 'your_token') in browser console (for downloads only, manually removed after use).
• Test uploads: Navigate pages and check predictpulse_realdata.json in predictpulse-data.
• Aggregate logs: GITHUB_TOKEN=your_token node scripts/upload_logs.js (server-side, after collecting logs).
• Use Vercel to securely store the GitHub token and request data from there, ensuring safe management and access to the private repo.

Privacy

User data is anonymized (page visits, device stats) and collected only with consent via a popup. Logs are stored in a private GitHub repo, accessible to collaborators.

License

MIT