🌌 ASTRAEA

Predictive Event Traffic Command Center for Bengaluru

---

📌 Executive Summary

Major public events—IPL Cricket Matches at M. Chinnaswamy Stadium, mega concerts at Bengaluru Palace Grounds, international exhibitions at BIEC, and massive rallies near Vidhana Soudha and Freedom Park—cause severe, cascading gridlock across Bengaluru’s arterial road networks. Traditional navigation apps operate reactively after congestion builds, leading to stranded commuters, trapped emergency vehicles, and delayed public transport.

Astraea is an end-to-end, AI-powered Predictive Event Traffic Command Center engineered specifically for urban traffic authorities in Bengaluru. Astraea moves traffic management from reactive firefighting to proactive dynamic dispatch.

By unifying Gradient-Boosted ensemble machine learning, spatiotemporal road network graphs (NetworkX), automated police deployment algorithms, and interactive geospatial visualization, Astraea enables traffic commanders to model event impacts hours in advance and deploy surgical interventions before gridlock occurs.

---

📸 System Showcase

1. Interactive Geospatial Command Dashboard

Real-time congestion heatmap, junction status overlays, venue footprint radius rings, and event simulation control panel.

---

2. Tactical Incident Response & Optimization

Automated officer budget allocation, dynamic signal green-split recommendations, and dynamic corridor diversion routing.

---

3. Post-Event Accuracy & Replay Analytics

Backtest model performance on real historical events (e.g. RCB vs CSK at Chinnaswamy Stadium) with MAE, RMSE, and actual-vs-predicted curves.

---

🔥 Key Features

• 🔮 Predictive Congestion Forecasting Engine: Utilizes a trained HistGradientBoostingRegressor model evaluated on over 575,000 spatiotemporal instances across 440+ historical and simulated events. Forecasts junction-level congestion index (0–100%) and vehicle delay (min/km).

• 🗺️ Bengaluru Road Graph Network (NetworkX): Models 38 key arterial junctions (e.g., Trinity Circle, Hebbal Flyover, Central Silk Board, Mekhri Circle, Corporation Circle) and 54 major connecting corridors with lane capacities, free-flow travel speeds, Haversine distances, and Betweenness Centrality metrics.

• 🚦 Automated Intervention & Dispatch Optimizer: Generates targeted tactical recommendations for traffic police commanders given a fixed manpower budget:

  • Officer Allocation: Prioritizes bottleneck nodes based on predicted surge and network centrality.
  • Signal Timing Adjustments: Recommends adaptive green-split extensions (+15s to +30s) on heavy corridors.
  • Dynamic Diversions: Identifies detour routes to protect central arterial nodes.

• ⏱️ Granular Spatiotemporal Timeline Scrubber: Interactive scrubber spanning T-3h (pre-event entry ramp-up) to T+3h (post-event dispersion), capturing arrival peaks, mid-event lulls, and departure surges.

• 📊 Post-Event Accuracy & Replay Analytics: Module for historical event backtesting (e.g. RCB vs CSK IPL matches, ColdPlay concerts). Computes MAE (Mean Absolute Error), RMSE (Root Mean Square Error), and percentage of predictions within ±5/10 point tolerances.

---

🏗️ Architecture & Data Flow

flowchart TD
    subgraph Inputs ["Input Layer"]
        A[Event Metadata: Venue, Attendance, Start Time, Duration]
        B[Weather & Calendar: Rain, Temp, Day of Week, Holiday]
        C[Road Graph Topology: 38 Junctions, 54 Corridors]
    end

    subgraph FeatureEng ["Feature Engineering"]
        D[Distance Decay Calculation]
        E[Temporal Offset Phase Mapping]
        F[Betweenness Centrality & Saturation Volumes]
    end

    subgraph ML ["Predictive ML Pipeline"]
        G[HistGradientBoosting Regressor]
        H[Congestion Index Prediction]
        I[Vehicle Delay Prediction]
    end

    subgraph Optimizer ["Intervention Engine"]
        J[Manpower Budget Allocator]
        K[Signal Timing Optimizer]
        L[Corridor Diversion Planner]
    end

    subgraph UI ["Interactive Command Center"]
        M[Leaflet Geospatial Map]
        N[Timeline Scrubber T-3h to T+3h]
        O[Tactical Action Cards]
        P[Historical Event Replay]
    end

    A --> D
    B --> E
    C --> F
    D & E & F --> G
    G --> H & I
    H & I --> J & K & L
    H & I & J & K & L --> M & N & O & P

Loading

---

🤖 ML Model Performance

The Astraea forecasting pipeline splits synthetic and historical event data by event ID to ensure strict out-of-sample evaluation on unseen public events.

Model Target	MAE (Mean Absolute Error)	R² Score	Baseline MAE	Skill vs. Baseline
Congestion Index (0–100)	3.049	0.9434	13.576	+77.5%
Vehicle Delay (min/km)	0.493	0.8185	1.066	+53.7%

Evaluated Features (17 Total)

• Spatial & Topology: dist_km, lanes, base_volume, centrality
• Event Properties: attendance, attendance_ratio, venue_capacity, event_type_code
• Temporal & Calendar: hour, dow, is_weekend, is_holiday, minutes_rel, duration_min, phase_code
• Environmental: rain, temp_c

---

📍 Bengaluru Venues & Junction Network

Modeled Venues (8 Curated Landmarks)

1. M. Chinnaswamy Stadium (Capacity: 40,000 | Cricket, Concerts)
2. Bengaluru Palace Grounds (Capacity: 100,000 | Concerts, Festivals, Rallies)
3. Freedom Park (Capacity: 50,000 | Political Rallies, Protests)
4. Vidhana Soudha Precinct (Capacity: 30,000 | VIP Movement, Rallies)
5. Sree Kanteerava Stadium (Capacity: 24,000 | Football, Athletics)
6. Kanteerava Indoor Stadium (Capacity: 12,000 | Indoor Sports, Concerts)
7. National College Grounds, Basavanagudi (Capacity: 35,000 | Cultural Festivals)
8. Jayamahal Palace Grounds (Capacity: 20,000 | Exhibitions, Expos)

Key Arterial Junctions (38 Total)

• Central Core: Trinity Circle, Anil Kumble Circle, Cubbon Park, MG Road, Residency Road, Richmond Circle, Corporation Circle, Town Hall, KR Market.
• Arterial Flyovers & Hubs: Central Silk Board, Hebbal Flyover, Mekhri Circle, Windsor Manor, Majestic (Kempegowda), Dairy Circle, Domlur Flyover, KR Puram / Tin Factory.

---

📁 Repository Structure

Astraea/
├── api/                        # Python FastAPI Backend
│   ├── app/
│   │   ├── artifacts/          # Trained model artifacts & metrics JSON
│   │   │   ├── model_congestion.joblib
│   │   │   ├── model_delay.joblib
│   │   │   └── metrics.json
│   │   ├── data/               # Spatial graph & venue definitions
│   │   │   ├── graph.py        # NetworkX road graph & centrality
│   │   │   └── venues.py       # Bengaluru venue database
│   │   ├── ml/                 # Machine Learning pipeline
│   │   │   ├── features.py     # Feature engineering & encoders
│   │   │   ├── forecast.py     # Forecaster wrapper class
│   │   │   ├── generate.py     # Historical event generator & actuals simulator
│   │   │   ├── optimize.py     # Police manpower & signal optimizer
│   │   │   └── train.py        # Training script
│   │   ├── main.py             # FastAPI entrypoint & REST routes
│   │   └── schemas.py          # Pydantic request/response schemas
│   ├── Procfile
│   └── requirements.txt
├── web/                        # Next.js 16 / React 19 Frontend
│   ├── src/
│   │   ├── app/                # Next.js App Router
│   │   │   ├── layout.tsx
│   │   │   └── page.tsx
│   │   ├── components/         # UI Components
│   │   │   ├── Dashboard.tsx               # Main Command Center Container
│   │   │   ├── MapView.tsx                 # Leaflet Map Component
│   │   │   ├── ScenarioPanel.tsx           # Scenario Input Controls
│   │   │   ├── TimeSlider.tsx              # Spatiotemporal Scrubber
│   │   │   ├── KpiBar.tsx                  # Top Metrics Display
│   │   │   ├── RecommendationsPanel.tsx    # Tactical Dispatch Cards
│   │   │   ├── AccuracyPanel.tsx           # Replay Evaluation Panel
│   │   │   └── TimelineChart.tsx           # Recharts Congestion Curve
│   │   └── lib/                # API client & types
│   ├── package.json
│   ├── tsconfig.json
│   └── postcss.config.mjs
└── docs/
    └── images/                 # Embedded visual assets
        ├── hero-banner.png
        ├── dashboard-preview.png
        ├── dispatch-optimization.png
        └── post-event-replay.png

---

⚡ Quickstart Guide

Prerequisites

• Node.js: v18.0.0 or higher
• Python: v3.10 or higher
• npm or pnpm

---

Step 1: Clone the Repository

git clone https://github.com/AdityaSahu786/Astraea.git
cd Astraea

---

Step 2: Set Up Backend (api)

1. Navigate to api/ directory:

   cd api

2. Create & activate Python virtual environment:

   python3 -m venv .venv
   source .venv/bin/activate    # On Windows: .venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Train the ML models:

   python -m app.ml.train

5. Start the FastAPI server:

   uvicorn app.main:app --reload --port 8000

   🌐 API Interactive Documentation: Available at http://localhost:8000/docs

---

Step 3: Set Up Frontend (web)

1. In a new terminal window, navigate to web/:

   cd web

2. Install Node dependencies:

   npm install

3. Start the Next.js development server:

   npm run dev

4. Open your browser and launch the Astraea Command Center:

   http://localhost:3000
   

---

📡 API Reference Overview

Endpoint	Method	Description
GET /health	GET	Health check & model artifact training status verification
GET /api/venues	GET	Returns all 8 Bengaluru venue records with coordinates & capacities
GET /api/event-types	GET	Returns event categories (cricket, concert, rally, festival, etc.)
GET /api/weather-options	GET	Returns rain/weather options for simulation
GET /api/graph	GET	Returns 38 graph nodes and 54 edges with centrality and road capacities
GET /api/metrics	GET	Returns model evaluation metrics (MAE, R², baseline comparison)
POST /api/simulate	POST	Generates junction-level forecasts and tactical intervention plans
GET /api/events	GET	Returns list of historical events available for replay
GET /api/events/{id}/replay	GET	Computes replay accuracy, predicted-vs-actual curves, and per-junction deltas

---

🗺️ Future Roadmap

• Live IoT Feed Integration: Ingest real-time traffic speeds from TomTom / Google Distance Matrix APIs and CCTV camera counts.
• V2X Signal Hardware API: Direct integration with Bengaluru Traffic Police (BTP) automated traffic signal controllers for dynamic green-wave pulsing.
• Reinforcement Learning Dispatcher: Deep Q-Network (DQN) agent for multi-agent officer routing under dynamic urban constraints.
• Public Transit Dynamic Rerouting: Dynamic shuttle frequency adjustment recommendations for BMTC buses during major event exits.

---

📜 License

Distributed under the MIT License. See LICENSE for more information.

---

Built with ❤️ for Bengaluru Urban Mobility & Traffic Command