Traffic Skipper

View Real-Time Traffic Conditions
Search for a Specific Highway

Inspiration

Traffic jams are not only leading to frustration, but also negatively impacting the environment. The mobile app provides a traffic forecast featuring a state-of-the-art machine learning model. This will hopefully make travelling the roads of the future faster, cheaper, safer and more ecological.

What it does

Loads transformed historical data from >800 sensors across Switzerland and tries to predict the next traffic flux states for each sensor. We also built a service (frontend+backend) to ease the access and visualisation of data.

How we built it

Using kaggle, we trained models that most appropriately fit our use case, monitoring the state. We are 5 team members, where two were fully dedicated into building the prediction model (ML team), another team member was exploring the mathematical approach to this problem, and the other two were developing the frontend and backend for the service.

Challenges we ran into

The provided data is quite sparse, so using continuous mathematical traffic models is not possible. 10Gb would take a lot of time to train the model, so we limited the training set to 1Gb which would yield less precise results at the end.

Accomplishments that we're proud of

Building and training an LSTM machine learning model.

What we learned

A lot about traffic theory, LSTM (ML) models (we experimented with pytorch and tensorflow), and not least important, team work.

What's next for Traffic Skipper

The first step to make Traffic Skipper successful would be to develop a more robust model with the complete data provided. Then, we would move on to make real time suggested.