BixiSense

Docking stations near Concordia University

Inspiration

Our inspiration for BixiSense came from the frustration of encountering misleading bike availability information on the Bixi app. We wanted to create an app that gives a user an overall idea of the odds that their plan of docking the bike or picking up a bike changes before they physically get to the station, especially when it comes to trip planning

What it does

BixiSense addresses the issue of unreliable bike availability information. The app displays a map of their nearby location, and a text field where they can enter a location, and see all the Bixi stations nearby. Depending on the mode they chose, they can either see the odds that dock availability or bike availability changes, based on parameters such as day of the week, month, time of day, and station location.

How we built it

We built BixiSense in two parts: data processsing/training, followed by app front-end implementation. Python, Pandas and ScikitLearn was used to process, clean and then train a model to predict the levels of activity in different Bixi stations, using the data sets made publicly available by Bixi. When it comes to our App, we used android along with the Google Maps API to integrate an intuitive, easy-to-use interface in which one can see the information that they want at a glance.

Challenges we ran into

During the development of BixiSense, we faced challenges such as the lacking dataset provided by Bixi, in which we had to make several assumptions, but also inspired us to think outside of the box. Additionally, implementing the Google Maps presented some difficulties, and as result we had to scrap some features that we initially wanted to include. The learning experiences gained from these challenges strengthened our skills and determination.

Accomplishments that we're proud of

Our main accomplishment that we're proud of is that we started several moving pieces (sub-tasks) and concepts that was going to be a hassle to merge all together and that despite the messy-ness of our workflow and hours of hard work, we managed to deliver a functional product to demo.

What we learned

Throughout the development of BixiSense, we gained valuable insights into the arts of creating an easy-to-use and intuitive interface for the user. Additionally, it gave us extra insight as to the precision and care needed to prepare data for training, as well as the need for thinking outside the box to extract meaning out of the data.

What's next for BixiSense

Looking ahead, we would like to implement some of the features that we had to scrap due to difficulties and time limitations. If the data was available, we would like to have a way to predict the estimated number of available docks/bikes given the same conditions we trained our model on, but this data was not present in the datasets. Additionally, we would like to have better weather integration, in which the model adjusts for weather conditions, and tweaking predictions. Finally, we would like to integrate Google Maps' routing integration, so that the app can also serve as an improved bixi trip planner.