🛡️ JATAYU - Predictive Intelligence Fusion System

Operation Silent Watch | Hackathon 2025

Anticipate IED threats before they materialize using multi-source intelligence fusion and machine learning.

---

🎯 Mission

Transform raw intelligence into actionable predictions. JATAYU analyzes historical IED incident data from India's Red Corridor to predict future attacks, enabling security forces to preemptively deploy resources.

---

📊 Real Data Statistics

Metric	Value
Total Incidents	187 (2020-2026)
States Covered	7 (CG, JH, OR, MH, TS, AP, Multi-border)
Total Killed	87
Total Injured	225
Top Hotspot	Bijapur (63 incidents, 27 killed)
Attack Clusters	39 detected

---

🚀 Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Run Complete Demo

python run_real_demo.py

3. Launch Dashboard

streamlit run src/visualization/real_dashboard.py

---

📂 Project Structure

jatayu_beta/
├── data/
│   ├── raw_incidents.csv      # Real IED incident data (187 incidents)
│   └── processed_incidents.csv
├── src/
│   ├── data/
│   │   ├── data_generator.py  # Synthetic data generator
│   │   └── real_data_loader.py # Real data loader & preprocessing
│   ├── features/
│   │   └── feature_engineer.py # Location-agnostic feature extraction
│   ├── models/
│   │   ├── models.py          # XGBoost + LSTM ensemble
│   │   ├── real_trainer.py    # Real data ML pipeline
│   │   └── explainer.py       # SHAP-based explanations
│   └── visualization/
│       └── real_dashboard.py  # Streamlit dashboard
├── run_real_demo.py           # Complete demo script
├── requirements.txt
└── README.md

---

🤖 ML Architecture

Features (23 engineered)

• Temporal: days_since_last, attack_velocity, attacks_last_7/30/90
• Spatial: district_historical_rate, is_high_risk_district
• Casualty: casualties_last_30, district_casualty_rate
• Seasonal: month_sin/cos, day_of_week_sin/cos

Model Performance

• Training: 139 samples (2020-2024)
• Testing: 48 samples (2025-2026)
• Accuracy: 58.3%
• Precision: 66.7%
• F1 Score: 0.667

Key Insights

1. Bijapur is the highest-risk district (63 incidents, 27 killed)
2. Attack tempo increases before major clusters
3. Casualty momentum is a strong predictive signal
4. Seasonal patterns detected (month_sin is #2 feature)

---

📈 Key Findings

Top 5 Hotspot Districts

1. Bijapur (Chhattisgarh): 63 incidents, 27 killed
2. West Singhbhum (Jharkhand): 33 incidents, 14 killed
3. Narayanpur (Chhattisgarh): 24 incidents, 13 killed
4. Dantewada (Chhattisgarh): 11 incidents, 12 killed
5. Sukma (Chhattisgarh): 11 incidents, 8 killed

Most Severe Attack Cluster

• April 2023: 4 attacks in 11 days, 11 killed, 1 injured
• Districts: Dantewada, Bijapur, Not specified

January 2025 Cluster (Validation Target)

• Jan 6-17, 2025: 6 attacks, 10 killed, 6 injured
• Districts: Bijapur, Sukma, Narayanpur, West Singhbhum

---

🔮 Current Risk Assessment

As of January 18, 2026:

• Attack Probability (7 days): 37.1%
• Risk Level: MEDIUM
• Recommended: Standard patrol protocols, continue monitoring

---

📚 Data Sources

The incident data was compiled from:

• South Asia Terrorism Portal (SATP)
• News archives
• Official reports

---

🛠️ Requirements

pandas>=1.5.0
numpy>=1.21.0
scikit-learn>=1.0.0
xgboost>=1.6.0
streamlit>=1.20.0
plotly>=5.0.0
folium>=0.14.0
streamlit-folium>=0.15.0

---

📝 License

For hackathon demonstration purposes only.

---

👥 Team

Built for Hackathon 2025 - Operation Silent Watch

---

Stay Vigilant. Predict. Protect.