Station Congestion Predicter

Inspiration

Every day, commuters and operators only react to congestion once platforms are already overflowing. We asked a simple question: what if we could see crowding coming days in advance, using data that already exists? That idea became the spark for a system that predicts station‑level congestion risk up to seven days ahead.

What it does

Station Congestion Predictor lets you pick any station and any date within the next week, and instantly see its expected congestion level. It displays a congestion index and a clear LOW / MED / HIGH risk label, learned from historical patterns rather than guesswork. Values above 100 indicate pressure worse than that station’s typical peak.

How we built it

We trained a multivariate machine‑learning model on historical station usage, weather forecasts, and calendar effects. The model learns how demand evolves over time and how external factors like rain or weekdays influence crowding. A clean ETL pipeline ingests raw public datasets and outputs a simple CSV that any interface can consume.

Challenges we ran into

Aligning multiple public datasets with different formats and time granularities was harder than expected. Station naming inconsistencies and missing values required careful cleaning. Embedding interactive maps in Streamlit without breaking tile layers also took some engineering finesse.

Accomplishments that we're proud of

We built a fully working end‑to‑end forecasting system that predicts congestion before it happens. The UI is simple, accessible, and decoupled from the model, making it easy to extend or reuse. Most importantly, the predictions are meaningful and interpretable for real commuters.

What we learned

We learned how to merge heterogeneous datasets into a coherent forecasting pipeline and validate a multivariate model on unseen data. We also learned how to translate ML outputs into something humans can understand at a glance. Ultimately, we realised prediction only matters when it’s actionable.

What's next for Station Congestion Predicter

We plan to add hour‑by‑hour forecasts, real‑time TfL integration, and personalised route suggestions. The same pipeline could scale to other cities or other transport modes with minimal changes. Our long‑term vision is to shift public transport from reacting to congestion to anticipating it.

Built With

• folium
• outlink
• python
• streamlit