The benefit of dockless transportation is that a user does not need to travel to a dock to start a journey. For docked bikes to have similar convenience to dockless vehicles, there would need to be more docks in high demand areas.

What it does

Our project will help to decide where to build more docks by predicting popularity of a dock based on current Bcycle usage. Our model would help us determine the amount of bikes that would be inbound and outbound of a new station based on factors such as proximity to other docks and elevation.

How we built

Every bike leaving a station has a probability of arriving at another station in the area. Using the dataset we calculated the probabilities of every trip and used this as data to train a random forest model. We can input the starting latitude and longitude of a new kiosk and the location of another kiosk to the movement of bikes towards and away from this station. If we calculate these probabilities towards and away from the new kiosk from every other kiosk, we can get useful information for kiosk placement.

Challenges we ran into

  • finding the ideal model to simulate probabilities of moving between nodes
  • hard to quantify important factors which affect kiosk popularity such as high traffic buildings and areas near the kiosk
  • removed known kiosks from training data and tested them later as "new" kiosks, but it is difficult to evaluate accuracy here as our only factors are location and elevation, and these are only one factor in how Bcycles are used.

What we learned

  • looking at graph data piece by piece can help create a big picture ## What's next for BCycle Kiosk Popularity Prediction Take in data from dockless transportation services such as Lime to further improve the model. This prediction could be used in a webapp that allows users to click on any point to "place" a new kiosk down and see it's predicted performance.

Built With

Share this project: