🚄 Predictive Transit Anomaly Detector

A full-stack application designed to monitor, predict, and visualize transit delays in real-time. By leveraging machine learning and live data streaming, this tool identifies potential "cascading warnings" and grid anomalies before they impact the entire network.

📋 Table of Contents

• Features
• Technologies Used
• Architecture
• Installation
• Usage
• Machine Learning Model

🚀 Features

• Real-Time Predictions: Continuously processes live transit data to predict station delays.
• Interactive Visualization: Dynamic Mapbox-powered dashboard showing station status across the San Francisco Bay Area.
• Anomaly Detection: Visual indicators for "Severe Delays" (>15 mins) with automated dashboard alerts.
• WebSocket Integration: Instant data broadcasting from the backend to the frontend for zero-latency updates.

🛠️ Technologies Used

• Frontend: React, Mapbox GL, React-Map-GL.
• Backend: Node.js, Express, WebSocket (ws).
• ML Engine: Python, XGBoost, Pandas, Scikit-learn.

🏗️ Architecture

The system consists of three primary components:

1. ML Engine: A Python-based XGBoost regressor that predicts actual_delay based on hour_of_day, temp, and prev_delay.
2. Node.js Backend: Acts as the orchestrator, simulating live data streams and spawning Python processes to run predictions.
3. React Frontend: A real-time map interface that receives updates via WebSockets and renders station health.

🔧 Installation

1. Prerequisites

• Node.js (v16+)
• Python 3.8+
• Mapbox Access Token (Free tier)

2. Setup ML Engine

cd ml-engine
pip install -r requirements.txt
python train.py # Generates the initial transit_model.json

3. Setup Backend

cd ../backend
npm install

4. Setup Frontend

cd ../frontend
npm install

💻 Usage

1. Start the Backend:

   cd backend
   node server.js

2. Start the Frontend:
   • Open frontend/src/App.js and replace YOUR_MAPBOX_ACCESS_TOKEN with your token.

   cd frontend
   npm start

3. View Results: Navigate to http://localhost:3000. The map will begin updating every 3 seconds with new simulated station predictions.

🧠 Machine Learning Model

The underlying model is an XGBoost Regressor trained on historical delay patterns. It considers environmental factors (temperature) and temporal factors (hour of day) to calculate the likelihood of a cascading delay.

• Input Features: hour_of_day, temp, prev_delay.
• Output: predicted_delay_mins.