Last month we were with our partners the City of Paris and Suez at the Paris city hall for a very special event. We presented the first results of our project about "Information about waste collection times" during the DataCity Demo Day. How to inform users, building caretakers and companies handling waste bins exactly when the waste collection truck will be driving by? That’s what we tried to solve.
46%, that's the reduction of bins presence in the street we promised, let's see what it means and how we'll fulfill it!
Rue de la Providence : poubelles, rétros, topcases = 70 cm de trottoir, impraticable en poussette. Paris piéton ? pic.twitter.com/6aIFbSP3PV— Un piéton véloce (@pieton_veloce) 24 février 2017
Almost 500 trucks pull out of waste processing facilities to collect rubbish on the streets of Paris every day, but bins cluttering public areas (pavements, entrance halls, etc.) are still a common sight, either because the trucks have not yet collected them or because nobody has returned them to storage.
The reason is simple: it is very hard to find reliable and precise sources of information about collection times. The only official source is the city website and what you’ll get there looks like that:
Routes are often updated due to works or organizational concerns. Bad weather conditions and traffic jams regularly cause delays. Those ever changing conditions make it extremely difficult for the people in charge of handling bins to plan for the right timing.
Paris city council regulations state that bins must not be taken outdoors more than an hour before collection and must be removed from the pavement 15 minutes after collection. But these rules are unfortunately hard to enforce without reliable information about when exactly the trucks will drive by.
DataCity is an open innovation program created by NUMA, a startup incubator, in partnership with the City of Paris. The objective is to use data to design solutions tailored for the benefit of tomorrow’s cities.
The objective is to make innovatives startups in technology collaborate with big french companies such as EDF, SFR, Suez, Setec, La Poste or Bouygues Energies & Services in order to solve urban problems using data to create the smart city of tomorrow.
As part of the second edition of DataCity, craft ai has been selected by the jury composed of the city of Paris and Suez for the challenge “Information about waste collection times”.
The main goal is to let the people in charge know when trucks will drive by their address in order to encourage them to put bins outside a few moment before arrival... and also bring them back in right after. Bins must stay as little time as possible outdoors. Benefits of such a service would be less obstruction on pavements in the street of Paris. That’s what the City of Paris is aiming at.
Direct users of this service would be those who are in charge of taking the bins outdoors, but information should be available to any citizen.
Waste collection data generated by waste trucks should be used for making predictions on time arrival at a given address in the 14th district of Paris (the area chosen for the challenge experimentation). The predictions should be accurate, but also the service should correctly deliver information to aimed people. The problem can be solved only if people use and act accordingly to predictions that have been made.
All bins in Paris have microchips that trucks read when they empty them. In the 14th district, there are 24 500 bins that all have this chip. Each bin is associated to an address, as shown in the diagram below. The City of Paris provided us with all the pickup event, that is when the bin is getting emptied, enriched with the truck name, its itinerary name, and its GPS position at pickup time.
All bins in the 14th district with chip embedded, visualized under CARTO
Moreover, there are two types of bins: the green ones (household waste) and the yellow ones (recyclable waste). In Paris, they are collected separately therefore the collection hour to predict might be different. This chip number identifies the bin color.
The data are collected since December of 2016 and month after month, the City of Paris is sending them to us.
The question was: how to predict, every day, the collection times for every address in the 14th district?
We used craft ai to learn collections times habits for both green and yellow bins. The entries variables for the models are the address, the time of day, the day of week and the type (green or yellow). The output variable is whether or not a waste truck is driving by the given address at given week day and time for given bin type. It resulted in craft ai agents who predict the waste truck arrival. Those agents learn everyday with collections data of the day. So the predictive model is continuously updated, hence fitting the latest habits. The system also take into account specific business rules in order to provide the most accurate and reliable predictions for people.
We measured on historical data how long would have stayed bins if people had used the predictions. We can state that it represents a 46% reduction of bins presence outdoors with very little risk of missing the collection. In other words, instead of 6 hours outdoors, we now provide the correct information that decrease it at less than 3 hours in all cases.
In the next article part, we will share more details about the complete workflow we used from data to predictions and the validation details!
Then, how to deliver it for concerned people?
We thought at what was the best way to inform people. A group of predictive models can’t be enough. We worked along with the City of Paris and Suez in order to define how citizens would access the predictions. It ended up with a planning online and a real-time service. Here is how it works.
Suez is hosting the predicted hours on monservicedechets.com. They are presented under a weekly planning. People from the 14th district will find their address and associated waste collection hours for both green and yellow bins.
Feel free to check it out for your own address and give us your feedback!
On the same page it’s also possible to subscribe to notifications. You just have to fill in your contact details, either phone number or mail. It works with three options you’re free to select.
Then what happens?
- You’re alerted 30 minutes before the predicted collection time at your own address, for both green and yellow bins. You’ll receive this kind of message: "Waste collection for green bins is estimated to take place at 8:45 at your location. Make sure to leave your bins out!". It’s like a reminder for what’s written on the website.
- You're alerted once the collection took place at your location. You’ll receive this kind of message: "Your green bins have been collected. Your can pull out them from the street.". Once the truck has read the chip your bin, we get this information in real time thanks to an API provided by Paris city and we send the notification to you.
- You’re alerted in case of incidents or perturbations related to waste collection. You’ll receive this kind of message in case: "Waste collection is canceled tomorrow at your location due to temporary waste truck issues". Paris waste management is behind this information. Instead of having your bin out the whole time, you can now be noticed early as possible for perturbations.
You can choose all of them or just one. The goal is to provide useful information to you!
Received notification about collection event at 1 Square Porte de Vanves of green bins
The DataCity challenge has recently come to an end and it was a real pleasure to work with the city of Paris, NUMA and Suez. We’ve met great people, dedicated to making Paris an even greater city.
The service is live for the 14th district of Paris, so if you happen to live there, go and give it a try, we’d love to hear your feedback.
We’re quite proud that craft ai approach to AI - individual level, continuous machine learning, whitebox models - helps fixing one of Paris old problem. The city has high ambitions for the service. We’re already working on improving predictions quality and preparing for a full scale deployment on every district of Paris. And we’re quite happy that other cities involved with the DataCity program have expressed interest in the service.
In the next article part, we will go Data Science mode and share insights about the workflow we implemented: data analysis, data augmentation, clustering and craft ai implementation. Stay tuned!