Road Health Map

Road Health Map is an admin dashboard for turning MPU6050 accelerometer readings into road-condition intelligence. The restored local version includes:

• a React authority dashboard with road-condition summaries, charts, a route visualization, feedback, tasks, profile, and settings;
• an Express API with token-based administrator authentication;
• MongoDB 7 in Docker with persistent local storage;
• automatic first-run import of all 10,023 classified readings from data/TotalData.csv.

The supplied CSV contains acceleration and class values but no GPS coordinates or timestamps. The sample surveyed route uses six field observations (three breakers and three rough patches) plus five midpoint-interpolated good-road points, producing 11 ordered markers. All markers remain attached to classified MongoDB sensor records, matching the original project data flow.

The frontend map provider is reversible through REACT_APP_MAP_PROVIDER: use osm for the free Leaflet/OpenStreetMap map, google for the preserved Google Maps component, or fallback for the built-in schematic. Restart the frontend after changing the value. Google mode additionally requires REACT_APP_GOOGLE_MAPS_API_KEY.

Run locally

Requirements: Docker Desktop, Node.js, and npm.

From this folder, start MongoDB:

npm run db:up

In a second terminal, start the API:

npm run backend

In a third terminal, start the frontend:

npm run frontend

Open http://localhost:3001 and sign in with:

• Email: admin@rhm.local
• Password: RoadHealth123!

The frontend uses port 3001 to avoid the existing Grafana service on port 3000. The API uses port 5050 because macOS commonly reserves port 5000 for Control Centre/AirPlay.

Configuration

The defaults work without local environment files. For custom values, copy backend/.env.example to backend/.env.

To enable the real Google map, copy frontend/.env.example to frontend/.env, add a browser-restricted Google Maps JavaScript API key, and restart the frontend. The frontend API URL can also be overridden with REACT_APP_API_URL.

Useful commands

npm run build       # production frontend build
npm run db:down     # stop MongoDB without deleting its data

MongoDB data remains in the rhm-mongo-data Docker volume. Starting the backend again does not duplicate seed records.

Rebuild the SVM model

The reproducible training script is ml/train_svm.py. It follows the archived implementation (RBF SVM, C=2, SMOTE, random state 2) and saves ml/svm_model.joblib.

The original Jupyter implementation is included in notebooks/Python Projects/SVM.ipynb, the collected source files are under data/ and sensor-values/, and portfolio-ready application images are available in screenshots/.

The verified model trained on all 10,023 rows with a stratified 80/20 split achieved 74.76% test accuracy and a 73.81% weighted F1 score.