🦟 DBDWaspada

AI-powered Dengue Hemorrhagic Fever Early Warning System
Platform Prediksi dan Peringatan Dini Wabah Demam Berdarah Dengue

About

DBDWaspada is an advanced AI-driven prediction system designed to forecast Dengue Hemorrhagic Fever (DHF/DBD) outbreaks at the kelurahan (sub-district) level. By integrating epidemiological data, weather patterns, satellite imagery, and social media sentiment, the platform provides a comprehensive early warning system for public health officials.

The system utilizes an XGBoost machine learning model trained on three years of weekly data (2022–2024) across 553 kelurahan in Jakarta, Bandung, and Surabaya. The model predicts outbreak risks two weeks in advance, allowing for proactive intervention. Risks are categorized into three tiers based on incidence rates: LOW (<20/100k), MEDIUM (20–55/100k), and HIGH (>55/100k).

Built specifically for Dinas Kesehatan (Health Departments), the dashboard offers interactive visualizations and explainable AI through SHAP values. This transparency helps decision-makers understand the specific environmental or social factors driving risk in each neighborhood, leading to more precise and effective preventative measures.

The platform streamlines the transition from data to action by automating multi-channel alerts. Whether it's rising temperatures or increasing social media reports of fever, DBDWaspada captures the signals early, helping cities move from reactive treatment to proactive prevention.

Screenshots

Peta Risiko (Risk Map)	Riwayat Peringatan (Alert History)

Monitoring Tweet	Navigation Sidebar

System Architecture

graph LR
    FE["🖥️ Frontend\nNext.js 15\n:3000"] -->|REST API| BE["⚡ Backend\nFastAPI\n:8001"]
    BE --> DB["🗄️ PostgreSQL 16\n+ TimescaleDB\n+ PostGIS\n:5433"]
    BE --> ML["📊 MLflow\n:5000"]
    BE --> PF["🔄 Prefect 3\n:4200"]
    style FE fill:#1e293b,color:#fff
    style BE fill:#0f766e,color:#fff
    style DB fill:#1d4ed8,color:#fff
    style ML fill:#7c3aed,color:#fff
    style PF fill:#b45309,color:#fff

Loading

Features

Feature	Description
🗺️ Interactive Choropleth Map	MapLibre GL JS + Deck.gl visualization at kelurahan level
🤖 XGBoost ML Prediction	2-week ahead dengue outbreak risk prediction
📊 SHAP Explainability	Per-kelurahan SHapley Additive exPlanations for model transparency
🔔 Multi-Channel Alerts	Automated warnings via WhatsApp, SMS, and Email
🌧️ Weather Integration	Rainfall, humidity, and temperature data
🛰️ Satellite Imagery	Standing water score from satellite data
🐦 Social Media Monitoring	Twitter/X sentiment analysis for DBD-related posts
📈 Trend Analytics	Historical charts with Recharts
🧪 MLflow Experiment Tracking	Full ML experiment reproducibility
🔄 Prefect 3 Pipeline	Automated data pipeline orchestration
📚 Swagger API Docs	Auto-generated interactive API documentation

Quick Start

# 1. Clone the repository
git clone https://github.com/vaskoyudha/DBDWaspada.git
cd DBDWaspada

# 2. Copy environment file
cp .env.example .env

# 3. Launch all services
docker compose up -d

# 4. Check service status
docker compose ps

Database Setup & Seeding

After the containers are running, initialize the database and generate the required mock data:

# Apply database migrations
docker compose exec backend alembic upgrade head

# Seed geographical data (Cities, Districts, Sub-districts)
docker compose exec backend python -m backend.seeders.kelurahan_seeder

# Generate mock datasets for each category
docker compose exec backend python -m backend.generators.dinkes_generator
docker compose exec backend python -m backend.generators.weather_generator
docker compose exec backend python -m backend.generators.satellite_generator
docker compose exec backend python -m backend.generators.tweet_generator

# Train the XGBoost model
docker compose exec backend python -m backend.ml.train

# Generate initial predictions
docker compose exec backend python -m backend.ml.predict

Service URLs

Service	URL	Description
🖥️ Frontend Dashboard	http://localhost:3000	Main UI
⚡ Backend API	http://localhost:8001	FastAPI REST API
📚 API Docs	http://localhost:8001/docs	Swagger UI
📊 MLflow UI	http://localhost:5000	ML Experiment Tracking
🔄 Prefect UI	http://localhost:4200	Pipeline Orchestration
🗄️ Database	localhost:5433	PostgreSQL (psql)

API Reference

All endpoints are prefixed with /api/v1.

Method	Endpoint	Description
GET	/api/v1/kota	List all cities
GET	/api/v1/kota/{kota_id}/kelurahan	Sub-districts per city (GeoJSON)
GET	/api/v1/kelurahan/{kelurahan_id}	Specific sub-district details (GeoJSON)
GET	/api/v1/predictions?city={name}	Risk predictions per city (GeoJSON + SHAP top-5)
GET	/api/v1/predictions/shap?city={name}&kelurahan_id={id}	SHAP waterfall values per sub-district
GET	/api/v1/alerts	List of alerts (filters, pagination)
GET	/api/v1/alerts/logs?channel={channel}	Alert delivery history logs
GET	/api/v1/alerts/{alert_id}	Specific alert details
POST	/api/v1/alerts/dispatch	Manually dispatch alerts for current predictions
GET	/health	System health check
GET	/docs	Interactive Swagger UI documentation

Tech Stack

• Backend: SQLAlchemy, Alembic, XGBoost, SHAP, Prefect 3, MLflow
• Frontend: Tailwind CSS v4, MapLibre GL JS, Deck.gl, Recharts
• Database: TimescaleDB, PostGIS
• Infrastructure:

Running Tests

Verify the system by running tests across the stack:

# Backend unit tests (Mocked DB)
docker compose exec backend python -m pytest backend/tests/ -m "not integration" -v

# Backend integration tests (Full DB)
docker compose exec backend python -m pytest backend/tests/ -v

# End-to-End pipeline tests
docker compose exec backend python -m pytest tests/e2e/ -v

# Frontend test suite
docker compose exec frontend pnpm test --run

Development Notes

• Network Ports: The database uses host port 5433 and the backend uses 8001 to avoid common local conflicts. Port 8000 and 5432 are reserved for internal container mapping.
• Reproducibility: Mock data generators use a fixed seed of 42.
• Epidemiological Weeks: Based on WIB (UTC+7) timezone.
• Data Coverage: 3 years of weekly data (2022–2024) across 7 city entities: Jakarta (5 administrative cities), Bandung, and Surabaya.
• Resolution: Covers 553 kelurahan: 262 in Jakarta, 137 in Bandung, and 154 in Surabaya.
• Risk Tiers:
  • LOW (RENDAH): < 20 per 100k population.
  • MEDIUM (SEDANG): 20 – 55 per 100k population.
  • HIGH (TINGGI): > 55 per 100k population.
• Prediction Target: The model predicts 2 weeks ahead (Target Week = Current Week + 2).

Contributing

Contributions are welcome! Please open an issue first to discuss what you would like to change. Pull requests should be against the main branch.

License

This project is licensed under the MIT License.