NextBus: Real-Time NYC Bus Arrival Time Prediction
Accurate, scalable bus ETA forecasting using spatio-temporal deep learning
Project Overview
// The Problem
Accurate bus arrival time prediction is challenging due to complex spatio-temporal dependencies, traffic congestion, irregular schedules, and large-scale streaming data. Traditional rule-based and statistical approaches struggle to capture non-linear interactions between routes, stops, and time-varying traffic conditions.
// The Solution
We developed NextBus, a real-time prediction framework that ingests live NYC MTA bus data and models both spatial and temporal dependencies using a Transformer-based spatio-temporal forecasting architecture. The system supports downstream stop forecasting and provides user-facing ETA predictions through a scalable web application.
// The Impact
NextBus improves ETA accuracy and user experience by reducing uncertainty in public transit planning. The project demonstrates how modern spatio-temporal deep learning models can be deployed end-to-end, from real-time data ingestion to user-facing prediction services, at city scale.
End-to-end real-time spatio-temporal forecasting system integrating live data ingestion, Transformer-based prediction models, and a web-based user interface.
Tech Stack
Key Features
- Real-time bus arrival time prediction for selected NYC routes and stops
- Downstream stop forecasting along bus trajectories
- Spatio-temporal modeling of bus movements and traffic patterns
- Scalable backend with real-time data ingestion
- User-friendly web interface for instant ETA queries
