As someone who’s passionate about making cities safer, I wanted to explore the relationship between collisions and location in London. I built a small app that uses machine learning to visualize live collision risk across the city. This isn’t just a static map – it’s an interactive tool that lets you see where and when collisions are more likely to happen.
The app uses historical TfL collision data and overlays risk on an interactive map. It’s powered by a simple tree-based classifier with probability calibration, so the scores are usable. I wanted to make it easy to discuss what features help, what doesn’t, and what’s misleading.
One of the key features of this app is its ability to show temporal risk scoring for London using a fixed spatial grid (H3 hexes) and time context. This means you can see how risk changes throughout the day and week.
I’m still iterating on the model, but I’ve already learned some interesting things. For example, I found that hour of day and day of week have a significant impact on collision risk. I’m also looking at adding external context like OSM history and weather to the pipeline.
If you’re interested in exploring the code, I’ve made it open source on GitHub. I’d love to hear your feedback and see how you use this app.
So, what do you think? Can we use machine learning to make our cities safer? I think this is a great start, and I’m excited to see where this project goes next.