🦟 Health.Env - Dengue Risk Prediction System

An intelligent machine learning system for predicting dengue outbreak risk levels across Indian cities using real-time environmental and demographic data. Features a modern glassmorphic dashboard with interactive data visualizations powered by an ensemble of ML models, with XGBoost achieving 96.57% accuracy.

---

📊 Overview

Health.Env combines a Flask REST API with multiple trained ML classifiers and a browser-based dashboard to forecast weekly dengue outbreak risk for 30+ major Indian cities. The system processes weather, environmental, and epidemiological indicators through a standardized preprocessing pipeline, returning probabilistic risk classifications (Low, Moderate, High) with confidence scores and actionable recommendations.

🏆 Model Comparison

We trained and evaluated 5 machine learning algorithms on 15,600 weekly dengue outbreak records (2015-2023, 30 Indian cities):

Rank	Model	Accuracy	Precision	Recall	F1-Score	Status
🥇	XGBoost	96.57%	96.75%	96.57%	96.59%	✅ Production Model
🥈	Random Forest	96.57%	96.75%	96.57%	96.59%	✅ Tied Accuracy
🥉	Decision Tree	96.44%	96.59%	96.44%	96.46%	✅ High Accuracy
4️⃣	K-Nearest Neighbors	95.32%	95.46%	95.32%	95.38%	✅ Instance-Based
5️⃣	Logistic Regression	89.68%	89.55%	89.68%	89.60%	⚠️ Baseline

Why XGBoost is Deployed:

• Tied highest accuracy with Random Forest at 96.57%
• Superior gradient boosting for sequential learning from errors
• Excellent feature importance interpretation for healthcare decisions
• Faster inference than Random Forest's ensemble of trees
• Better generalization on unseen data due to L1/L2 regularization
• Handles class imbalance effectively with scale_pos_weight
• Industry-standard for healthcare ML applications

Model Selection Rationale: While XGBoost and Random Forest achieve identical accuracy (96.57%), we chose XGBoost as the primary production model because:

1. Interpretability: Built-in feature importance (gain, cover, weight) for medical professionals
2. Performance: Gradient boosting sequentially corrects prediction errors
3. Scalability: Single boosted tree vs 100+ trees in Random Forest
4. Robustness: Regularization (gamma, lambda, alpha) prevents overfitting on seasonal patterns
5. Speed: Faster training (0.94s vs 0.84s) but more efficient inference

KNN Model: A K-Nearest Neighbors model is also trained and available for predictions. It uses distance-weighted voting to classify risk levels based on similarity to historical cases, making it useful for:

• Explainable predictions ("similar to outbreak X in city Y")
• Local pattern recognition
• Cases where instance-based reasoning is preferred
• Comparison with tree-based ensemble methods

The API serves predictions from all 5 models simultaneously (XGBoost, Random Forest, Decision Tree, Logistic Regression, KNN), allowing users to:

• Compare results across different algorithms
• Choose based on specific requirements (speed vs accuracy, interpretability vs performance)
• Ensemble predictions for higher confidence
• Analyze model agreement for risk assessment

---

✨ Key Features

🤖 Machine Learning

• 5 Trained Models: XGBoost, Random Forest, Decision Tree, K-Nearest Neighbors, Logistic Regression
• All 5 officially evaluated with comprehensive comparison metrics
• XGBoost as primary production model (96.57% accuracy, tied with Random Forest)
• KNN provides instance-based predictions with 95.32% accuracy
• Trained on 15,600+ weekly historical records (2015-2023)
• StandardScaler preprocessing for normalized predictions
• Multi-class classification: Low, Moderate, High risk levels
• Real-time predictions via REST API endpoints
• All models available for prediction and comparison

📊 Dual Dashboard System

1. Main Dashboard (/)

Real-time dengue risk prediction interface:

• Interactive input controls for 12 environmental/demographic parameters
• Live prediction with confidence scores and risk visualization
• Risk-specific actionable recommendations
• 4 interactive Plotly charts displaying real training data:
  • Monthly Risk Trend (seasonal patterns)
  • Rainfall vs Temperature scatter (risk-colored)
  • Feature Importance from models
  • City Risk Distribution across India
• Theme toggle (dark/light mode)
• Offline fallback mode

2. Model Comparison Page (/compare)

Comprehensive ML model analysis dashboard:

• 🏆 Best Model Card: Highlights XGBoost with all metrics (96.57% accuracy)
• 📊 Accuracy Bar Chart: Visual comparison of all 5 models
• 🎯 Radar Chart: Multi-dimensional performance view
• 📈 Grouped Bar Chart: Side-by-side metric comparison
• 📋 Performance Table: Detailed metrics with rankings
• 💡 Insights Section: AI-generated analysis of model strengths

🎨 Modern UI/UX

• Glassmorphic design with smooth animations
• Fully responsive (desktop → tablet → mobile)
• Dark/light theme with localStorage persistence
• Interactive charts with zoom and export
• Navigation between dashboard and comparison pages
• Real data from 8 years of training history

---

🚀 Quick Start

Prerequisites

• Python 3.8+ (Python 3.11 recommended)
• pip package manager
• Modern web browser

Installation

# Clone repository
git clone https://github.com/Chanu716/Health.env.git
cd Health.env

# Create virtual environment
python -m venv .venv
.\.venv\Scripts\Activate.ps1  # Windows
# source .venv/bin/activate    # macOS/Linux

# Install dependencies
pip install -r requirements.txt

Running Locally

# Start Flask API
python api.py

# Access dashboards
# Main: http://localhost:5000/
# Comparison: http://localhost:5000/compare

---

🔌 API Endpoints

Endpoint	Method	Description
/	GET	Main prediction dashboard
/compare	GET	Model comparison page
/health	GET	API health check
/predict	POST	Get dengue risk predictions (all models)
/model-info	GET	Model metadata
/api/feature-importance	GET	Feature weights
/api/training-stats	GET	Training data statistics
/api/model-comparison	GET	All model metrics

Example: POST /predict

Request:

{
  "city": "Mumbai",
  "month": "July",
  "temperature": 32.5,
  "rainfall": 1200,
  "humidity": 85,
  "aqi": 150,
  "mosquito": 0.75,
  "population": 5000,
  "cases": 450
}

Response:

{
  "success": true,
  "predictions": {
    "xgboost_model": {
      "risk_level": "High",
      "confidence": 94.23,
      "probabilities": {"low": 1.42, "moderate": 4.35, "high": 94.23}
    },
    "random_forest_model": {
      "risk_level": "High",
      "confidence": 93.87,
      "probabilities": {"low": 1.58, "moderate": 4.55, "high": 93.87}
    },
    "decision_tree_model": {
      "risk_level": "High",
      "confidence": 91.20,
      "probabilities": {"low": 2.10, "moderate": 6.70, "high": 91.20}
    },
    "knn_dengue_model": {
      "risk_level": "High",
      "confidence": 88.50,
      "probabilities": {"low": 3.20, "moderate": 8.30, "high": 88.50}
    },
    "logistic_regression_model": {
      "risk_level": "Moderate",
      "confidence": 52.30,
      "probabilities": {"low": 15.40, "moderate": 52.30, "high": 32.30}
    }
  }
}

---

📁 Project Structure

Health.env/
├── api.py                          # Flask REST API
├── templates/
│   ├── index.html                  # Main dashboard
│   └── compare.html                # Model comparison
├── static/
│   ├── script.js                   # Dashboard logic
│   ├── compare.js                  # Comparison page logic
│   └── styles.css                  # Styling
├── models/
│   ├── xgboost_model.pkl           # 96.57% (primary production)
│   ├── random_forest_model.pkl     # 96.57% (tied accuracy)
│   ├── decision_tree_model.pkl     # 96.44% (fastest)
│   ├── knn_dengue_model.pkl        # KNN (instance-based)
│   ├── logistic_regression_model.pkl # 89.68% (baseline)
│   ├── scaler.pkl                  # StandardScaler (15 features)
│   ├── feature_names.pkl           # Core 12 features
│   └── feature_names_15.pkl        # Extended 15 features
├── data/
│   ├── dengue_data_cleaned.csv
│   └── dengue_india_weekly_with_nulls.csv
├── results/
│   └── model_comparison_results.csv
└── requirements.txt

---

🧪 Model Details

Features (12 Core Variables)

1. Year & Week (temporal)
2. Temperature (°C)
3. Rainfall (mm)
4. Humidity (%)
5. Air Quality Index (AQI)
6. Mosquito Density (0-1)
7. Population Density
8. Dengue Cases Reported
9. Latitude & Longitude
10. Month Number

Training Configuration

• Dataset: 15,600 records (2015-2023)
• Split: 80-20 stratified
• Preprocessing: StandardScaler normalization
• Validation: 5-fold cross-validation
• Tuning: GridSearchCV for hyperparameters

Supported Cities (30)

Mumbai, Delhi, Bangalore, Hyderabad, Chennai, Kolkata, Pune, Ahmedabad, Jaipur, Lucknow, Kanpur, Nagpur, Indore, Thane, Bhopal, Visakhapatnam, Pimpri-Chinchwad, Patna, Vadodara, Ghaziabad, Ludhiana, Agra, Nashik, Faridabad, Meerut, Rajkot, Kalyan-Dombivli, Vasai-Virar, Varanasi, Srinagar

---

🤝 Contributing

Contributions welcome! Areas for improvement:

• Add LSTM/GRU for time-series forecasting
• Implement geographic heatmaps
• Mobile app (React Native/Flutter)
• Real-time data integration
• Multi-language support

---

📄 License

MIT License - See LICENSE file

---

👨‍💻 Author

Karri Chanikya Sri Hari Narayana Dattu

• GitHub: @Chanu716
• Project: Health.env

---

🙏 Acknowledgments

• Scikit-learn & XGBoost teams
• Flask & Plotly.js communities
• Indian health departments for data
• Open-source ML community

---

Built with ❤️ for public health | Preventing dengue outbreaks through AI

Version 2.0 | December 2025