🚆 Rail Drishti — Indian Railways Intelligence System

"Drishti" (दृष्टि) means Vision in Hindi. Rail Drishti gives you intelligent, data-driven vision into India's railway network.

---

📌 Project Overview

Rail Drishti is a full-stack AI-powered railway intelligence platform built on top of real Indian Railways data. It provides two distinct dashboards — one for Passengers and one for Railway Admins — each powered by Machine Learning models running on a FastAPI backend, with a sleek dark-themed frontend served via Vercel.

The system ingests 4 real datasets (stations, schedules, train details, historical delay data), merges them into a unified master dataset, trains Random Forest models in the background on startup, and exposes a clean REST API consumed by the frontend.

---

🗂️ Project Structure

databrick-main/
├── backend/
│   ├── app.py              # FastAPI server — all API routes
│   ├── models.py           # ML models: DelayPredictor, CongestionClassifier + analytics
│   ├── data_loader.py      # Data ingestion, cleaning, and master dataset builder
│   ├── utils.py            # Terminal/ANSI display utilities
│   └── requirements.txt    # Python dependencies
├── frontend/
│   ├── index.html          # Single-page app (Passenger + Admin tabs)
│   └── config.js           # API base URL config
├── data/
│   ├── stations.json           # GeoJSON — Indian railway stations with coords/zone/state
│   ├── schedules.json          # Train schedules per station
│   ├── Train_details_22122017.csv  # Train routes, stop sequences, timings
│   └── train_delay_data_rich.csv   # Historical delay data (ML training set)
├── Dockerfile
├── docker-compose.yml
└── railway.json            # Railway.app deployment config

---

⚙️ Tech Stack

Layer	Technology
Backend	Python 3, FastAPI, Uvicorn
ML	scikit-learn (RandomForestRegressor, RandomForestClassifier)
Data	Pandas, NumPy, joblib
Frontend	Vanilla HTML/CSS/JS (no framework)
Deployment	Railway (backend), Vercel (frontend), Docker

---

🚀 Getting Started

Prerequisites

• Python 3.10+
• pip

Run Locally

# 1. Install dependencies
cd backend
pip install -r requirements.txt

# 2. Start the server
uvicorn app:app --host 0.0.0.0 --port 8000 --reload

# 3. Open frontend
open frontend/index.html
# OR set VITE_API_URL in frontend/config.js and serve via any static server

Docker

docker-compose up --build

Environment Variables

Variable	Description	Default
PORT	Server port	8000
RAILWAYS_DATA_DIR	Path to data folder	./data
FRONTEND_URL	Allowed CORS origin for your Vercel frontend	(none)

---

🤖 Machine Learning Models

1. DelayPredictor — Random Forest Regressor

Predicts arrival delay in minutes given route features.

Input features:

• distance_km — distance between stations (clamped to 0–955 km)
• weather — Clear / Rainy / Foggy / Stormy / Hazy
• day_of_week — Monday to Sunday
• time_of_day — Morning / Afternoon / Evening / Night / Late Night
• train_type — Rajdhani / Shatabdi / Duronto / Superfast / Express / Local / Passenger

Output: Delay in minutes, scaled per train type:

• Rajdhani/Shatabdi: capped at 25–30 min
• Duronto: capped at 45 min
• Express/Superfast: capped at 75–90 min
• Local/Passenger: capped at 50–60 min

Config: 50 estimators, max_depth=10, min_samples_leaf=5

---

2. CongestionClassifier — Random Forest Classifier

Classifies a route as Low / Medium / High congestion.

Same 5 input features as DelayPredictor. Also supports predict_proba() to return probability breakdown per class.

Config: 50 estimators, max_depth=8

---

🎫 Passenger Dashboard Features

U1 — Train Search

Enter any train number to fetch complete intelligence for that train:

• Train name, type, source, destination, distance
• Predicted delay (ML-powered, in minutes)
• Congestion level of the route (Low / Medium / High)
• Risk level (Low / Medium / High)
• Reliability score (0–100)
• Current weather context
• Time of day bucket

U2 — Delay Reasoning Engine

After predicting delay, the system generates human-readable reasons for it:

• 🌫️ Foggy conditions → reduced speed operations active
• 🌧️ Rainfall → track adhesion and signal visibility affected
• ⛈️ Storm advisory → speed restrictions imposed
• 🔴 Route heavily congested → multiple trains sharing track
• ⏱️ Peak hour (Morning/Evening) → signal queuing likely
• ⚠️ Cascading upstream delay
• 🚨 Severe delay → possible track maintenance

U3 — Reliability Score

A 0–100 score calculated from:

• Number of stops (complexity penalty)
• Average route distance (exposure penalty)
• Median historical delay (delay penalty)

Displayed as both a number and a visual progress bar.

U4 — Stop-by-Stop Schedule with Estimated Arrival

For each stop on the route (up to 15 shown):

• Station code + name
• Scheduled arrival and departure times
• Estimated arrival = scheduled time + predicted delay minutes
• Distance in km
• Platform number (where available)

U5 — Alternative Trains on Same Route

Shows other trains that share at least 2 stations with the searched train — useful as alternatives when the main train is delayed.

---

🏛️ Admin Control Dashboard Features

A0 — Operational Dashboard (Summary KPIs)

Quick-load overview:

• Total trains in network
• Total stations
• Total route entries
• Zone-wise train distribution

A1 — Most Congested Stations

Top-N stations ranked by congestion score:

congestion_score = trains_stopping × log(1 + routes_through)

Displays station code, name, zone, state, train count, route count, visual gauge bar, and Low/Medium/High badge.

A2 — Congested Rail Corridors

Top-N route corridors (Source → Destination) ranked by route score:

route_score = trains_on_corridor × log(1 + stations_on_corridor)

Includes corridor name, train count, station count, max distance, and load gauge.

A3 — Congestion Risk Forecasting (Multi-Window)

Predicts congestion risk across 3 time windows:

• Now — current conditions
• +2 Hours — near-future
• Tomorrow AM — next-day morning

For each window, runs 4 scenario predictions (Express/Superfast/Local/Foggy-Express) with predicted delay and congestion level.

A4 — Cascading Delay Simulation

Given a train number, simulates how its delay propagates to downstream trains:

• Analyses top route corridors for that train
• Predicts base delay per corridor (ML model)
• Applies a 1.4× cascade factor (each downstream stop picks up ~20–40% more delay)
• Classifies cascade risk as Low / Medium / High
• Returns: input train, affected trains, station hubs, delay passed in minutes

A5 — Rerouting Options

For a given train, finds alternative trains sharing ≥ 2 stations with it, ranked by shared station count. Returns top 5 with train number, name, route, and type.

A6 — Priority Clearance Queue

Lists trains sorted by operational priority:

• P1 — Rajdhani, Shatabdi, Duronto (highest priority)
• P2 — Superfast
• P3 — Express
• P4 — Local, Passenger

Used to decide which trains get signal/track clearance first in a congestion scenario.

A7 — Zone / Division Summary

Shows all Indian railway zones with:

• Number of stations
• Number of trains
• Number of states covered
• Overall congestion level (Low / Medium / High)

A8 — Network Bottleneck Detection

Automatically identifies stations with extreme congestion:

• Flags stations above the 90th percentile congestion score threshold
• Classifies them as: High / Critical / Extreme
• Provides reason: "Congestion score above 90th percentile threshold"

---

🌐 API Reference

System

Endpoint	Description
GET /	Health ping
GET /health	Full status: loaded, train count, station count

Passenger (User) APIs

Endpoint	Params	Description
GET /api/user/train-info	train_no	Full train intelligence
GET /api/user/alternatives	train_no	Alternative trains on same route
GET /api/user/station-info	station_code	Station metadata
GET /api/user/nearby-stations	station_code, radius_km	Haversine-based nearby stations

Admin APIs

Endpoint	Params	Description
GET /api/admin/congested-stations	top_n	Congestion-ranked stations
GET /api/admin/congested-corridors	top_n	Congestion-ranked corridors
GET /api/admin/congestion-risk	—	Multi-window risk forecast
GET /api/admin/cascading-delays	train_no	Cascade simulation
GET /api/admin/rerouting	train_no	Rerouting options
GET /api/admin/zone-summary	—	Zone-level operational summary
GET /api/admin/bottlenecks	—	Network bottleneck stations
GET /api/admin/priority-clearance	top_n	Priority-sorted train queue
GET /api/admin/operational-dashboard	—	Full KPI summary

---

📊 Data Sources

File	Records	Description
stations.json	~8,000+ stations	GeoJSON with lat/lon, zone, state
schedules.json	~100,000+ entries	Arrival/departure times per stop
Train_details_22122017.csv	~200,000+ rows	Route sequences with distances & timings
train_delay_data_rich.csv	~10,000+ rows	Historical delay data (ML training set)

---

🎨 Frontend Design System

The UI uses a dark-mode design with custom CSS variables:

Token	Value	Use
--bg-base	#05090f	Page background
--accent-blue	#1a8fff	Primary actions
--accent-cyan	#00d4ff	Highlights
--accent-amber	#f59e0b	Warnings
--accent-red	#ef4444	Danger/high delay
--accent-green	#22c55e	OK/low delay
--font-display	Bebas Neue	Headers
--font-mono	Space Mono	Codes/data
--font-body	DM Sans	Body text

Additional UX features: noise texture overlay, animated pulse status dot, live clock ticker, gradient glows on cards, sticky header with tab switching.

---

🚢 Deployment

Backend → Railway.app

// railway.json
{ "build": { "builder": "DOCKERFILE" }, "deploy": { "startCommand": "uvicorn backend.app:app ..." } }

Frontend → Vercel

// vercel.json
{ "rewrites": [{ "source": "/(.*)", "destination": "/index.html" }] }

Set FRONTEND_URL env var on Railway to your Vercel deployment URL to enable CORS.

---

📄 License

This project uses publicly available Indian Railways data. Built for educational and research purposes.

---

Rail Drishti — See the railway, understand it, optimise it.