Bluebikes Demand Forecasting

View live station data

Inspiration

Boston’s Bluebikes network is everywhere — but anyone who rides often knows the frustration of arriving at a station with no bikes or no docks. We wanted to build a tool that could predict availability before it happens, using real data and live feeds. Our inspiration came from everyday inconvenience and the question: “Can we see the future of bike demand — even just one hour ahead?”

What it does

Bluebikes Demand Forecasting is a real-time web app that:

Pulls live data from Bluebikes’ public GBFS API every few minutes
Predicts bike and dock availability for the next 60 minutes at each station
Visualizes it all on an interactive map, color-coded by predicted shortage or surplus

It helps riders plan smarter and lets planners instantly spot high-demand zones.

How we built it

We split the project into three quick but powerful parts:

Data Ingestion: Collected live station data from the GBFS API and historical trip data to compute short-term patterns.
Modeling: Built a lightweight regression-based predictor (linear/XGBoost) to forecast the number of available bikes at each station in one hour. $$\hat{y}{t+60} = f(\text{bikes}{t}, \text{docks}_{t}, \text{hour}, \text{weekday})$$
Frontend: Developed a Next.js + Leaflet interactive map with real-time updates, connected to a Flask backend serving predictions. We deployed both locally using Docker for easy integration.

Challenges we ran into

Combining real-time streaming data with a machine learning model in only 30 hours
Debugging Leaflet re-rendering issues in React (hello, “Map container already initialized”)
Coordinating between Python (Flask) and TypeScript (Next.js) under tight time pressure
Managing API rate limits and caching live station snapshots

Accomplishments that we're proud of

Built a fully functional end-to-end system in just 30 hours
Integrated real-time data + prediction + visualization seamlessly
Achieved smooth live updates and clear visual indicators for risk levels
Learned how to deploy and connect cross-language services efficiently

What we learned

How to design a data pipeline under extreme time constraints
The importance of lightweight, interpretable models in real-time systems
How to bridge machine learning with live web technologies (Python ↔ Node ↔ React)
That even a small forecast window (60 min) can make city mobility smarter

What's next for Bluebikes Demand Forecasting

Integrate weather and event data to improve accuracy
Deploy a public web dashboard for Boston riders
Extend to 15-min rolling forecasts and uncertainty intervals
Eventually help the city with bike rebalancing insights and sustainability planning