We looked over some Twitter feedback for the railway companies. The major problem we identified is that the answer is usually the very silly one: "Which train are you in?". Not only that most people don't ever reply again, but if they do, most often they don't provide the exact information which would enable decision makers to identify the exact train service. Besides wasting time in which action could be taken, this is by far a very bad customer experience. We want to solve this. Our solution uses advanced prediction algorithms to match a Tweet with the train it was sent from. This enables the railway companies to take action possibly even before people who made the complaint get off the train. Imagine how valuable this saved time is if the tweet actually reports a major incident which could put the lives of the passengers in danger.
What it does
It maps time and location to the train service that time and location corresponds to (if any).
How I built it
Our algorithm works as follow:
- Given a geolocation, a date and a time, the algorithm looks around for stations on a radius of 30km.
- We analyse all the trains passing through those stations in an interval of +- 2 hours.
- We fetch the trajectories of those trains formed of points 10 meters apart.
- We filter those trains which have a point on their trajectory which is very close to our geolocation.
- For these points which match, using interpolation between station times and locations, we predict the time the train would be in any of those points.
- We will choose the rail service which corresponds to the point which also matches with the provided time.
Challenges I ran into
Understanding the datasets and combining multiple datasets in a reliable way.
Accomplishments that I'm proud of
We manage to predict with a reliable degree of accuracy the rail services which pass through that point.