Exactly one year ago, we have closed our parking data pilot for Helsinki City in the Katajanokka area. In this blog post we go into details about what we’ve learnt and how we will use such learnings for our future services. Read on.
In Spring 2016 we’ve been working with Helsinki City to test and prove our capability to work with parking data collected from drivers. A sample of residents in the Katajanokka area got a device that allowed us to track their parking behaviour. During the project we acquired loads of learnings about how to use anonymised driver-level parking data to tell drivers where it is more likely they can find available on-street parking spaces.
The pilot was extremely positive and it gave Helsinki City good insights about what can be achieved in terms of parking analytics and occupancy predictions. Here is how Lauri Uski, Project Manager for Helsinki City, commented on the Katajanokka pilot.
Helsinki City got important information about the moving patterns of drivers. At the same time, parking behaviour predictions have been valuable to understand how the area can be developed in the future to meet parking needs. The open communication with drivers to get feedback and better understand their parking pain points has also proved extremely insightful.Lauri Uski
During the pilot we have achieved on-street parking occupancy prediction accuracy of 90%. We will anyway focus today on anonymised driver-level parking data, as we’ve gone through latest successes with our predicting algorithm in a recent blog post.
Driver-level Parking Data
In the picture below you can see an extremely simplified version of the information we could get. You can see three example cars with a fairly predictable behaviour during working days. Since we were getting information from Katajanokka residents, we could track their car parking behaviour for few weeks to learn and inform our model.
Car one leaves its parking space at 6 and is back at 13; car two leaves its parking space at 8 and is back at 16; car three leaves its parking space at 11 and is back at 15. The drivers had fairly consistent behaviour throughout the pilot, also in terms of where they could park.
How can we use this? Should a fourth driver want to visit area 4 on a Friday, it would be possible for us to recommend it would visit between 11 and 13, when most of the cars are out of the area. At the same time, should a driver be in the proximity of the area at 10, it would be possible for us to direct her in the area close to Kanavakatu and Laukkasaarenkatu, as she would have more possibilities to find an available parking space there (drivers represented by car 3 would still be parked in Merikasarminkatu).
We’d like to stress once again that this is an over-simplification to explain how we can use anonymised driver-level parking data in a way that makes car parking a positive experience and a sustainable activity. In this particular example, we were tracking way more than three cars in the area. And of course not all cars in the area are owned by residents or have such a regular behaviour.
Yet our estimates for car parking occupancy in area 4 achieved an accuracy of above 90% during the one month the pilot was on. This was a pretty remarkable success for us, and gave us confidence to continue our research in this direction.
Putting it all together
In this second picture, you can see how we could aggregate all anonymised driver-level parking data to give reliable guidance to drivers in search of a parking space.
In this case, we look at a specific date and time (Thursday May 5th 2016, at 13.00). Should a driver be close to Katajanokka by the time, we could suggest the areas where she could most likely find an available car parking space depending on which area she is going to visit. For example, if she would have to go to the Cathedral or to have a coffee at Signora Delizia, both Areas 1 and 3 would be a good choice. On the other hand, if she would be going to the Eastern part of the island, we could advise to stop for parking in Area 3 or 4, instead of going all the way to Area 5.
What’s in a year
As said, the Katajanokka pilot ended one year ago. The work with Helsinki City and the results we achieved made us confident that what we were working on back then was extremely relevant. And could help us change the way we park.
In one year, we have further explored different data sources to improve car parking availability predictions and we have worked to refine our internal capabilities. We know we can tell drivers where to park their cars, and we’ll soon unveil a service that will eventually aim at doing exactly that.