Show and Tell - Predict Passenger Frequencies

Hi Folks,

It‘s not brand new but I’d like to share it anyway: I made a Dash App to analyze and predict passenger frequencies at the bus and tram station Hardbrücke in Zurich:
https://hardbruecke.herokuapp.com/

It’s in German but quite easy to understand: You can choose any day and entrance to the station via the radio buttons. The graph shows the historical data of that day in blue (if available) an the predictions in red. The value is the number of people who entered or left the station at that time. The predictions are made by a simple random forest model.

In the tab “Historische Daten” you can analyze the historical data over time (week, month, quarter, day) by various aggregations (mean, min, max, …).

In the tab “Situationsplan” you can find general information such an pictures that show the position of the different entrances.
You can find the code here:

I hope you like and look forward to read you feedback.

Cheers
Alex

2 Likes

Nice app, Alex @alexander.guentert
I was able to use Google translate on the page and understood the content even better :slight_smile:

That historical data tab is pretty cool. What gave you the idea of making this app? Is this for work?

And thank you for making the code public :pray:

1 Like

Nice work! Didn’t expect many Swiss people here, but we’re already two now :wink:

2 Likes

Wow, that was fast. Thank you Adam!
I was just browsing the open data website of the city of Zurich and that dataset caught my attention. I am working in the public transport sector and knew that these kind of data often show regularities so my guess was that these frequencies can be predicted. But I did not do that for work it’s a pure fun an practice project.
In fact the machine learning and building a dash app part was pretty straight forward. What I struggled the most with, was heroku. In the beginning I had a pretty big model and downloaded the whole historical data. But heroku seemed to have some restrictions in case of RAM. So I had to downsize the model and use the api from that open data portal, which worked out pretty well in the end.

pretty nice solution you found for that heroku challenge. I’ve faced the same challenge, so I might use your solution :slight_smile: