For best experience please turn on javascript and use a modern browser!
You are using a browser that is no longer supported by Microsoft. Please upgrade your browser. The site may not present itself correctly if you continue browsing.

Maarten Sukel wants to improve the liveability of cities. To achieve this, he is working on machine learning models with various time frames. ‘For the city, I’m trying to solve day-to-day problems happening right now…. at the UvA, I’m also investigating how one can keep this kind of problem from happening in the future.’

Machine learning

Every year, the city of Amsterdam receives over 250,000 service requests involving the public space. These requests, which can be submitted online or by telephone, are very diverse. There may for example be an old couch on the pavement, night owls may keep city centre residents awake or a speedboat may zip by swimmers on the Amstel river. Until recently, all requests were categorised manually, which could produce incorrect categorisations if jargon was misunderstood. Since late 2018, the city has been using machine learning to better classify service requests, explains Maarten Sukel, a PhD student at the Amsterdam Business School (ABS), which is part of the University of Amsterdam (UvA).

Sukel was one of the first master’s students in Data Science who did his PhD research at the Municipality of Amsterdam. He developed a text classification model that can automatically classify requests into one of eight categories and over seventy subcategories so that appropriate action can be taken right away. This model was implemented immediately for requests concerning public areas in Amsterdam. Sukel used a municipal database with more than 500,000 historical requests and complaints. ‘Previously, a couch might be classified as household rubbish, and a garbage-bag collecting truck would be dispatched – thus wasting several days and resulting in additional costs and a disappointed citizen. Or someone might go take a look during the day on a Tuesday in a street that had been reported as very noisy in the night of Friday to Saturday - and not hear anything.’ Thanks to the classification model, this problem is being dealt with. ‘The first version is ready. While it works, there is still room for improvement. ‘There are deep learning techniques that work better. In addition to text, it’s also possible to include locations, images and a timestamp in the classification.’

‘At the city, I’m trying to solve day-to-day problems happening right now.'

A first, with many more to follow

Sukel has been working on his PhD research for the past year and a half as a PhD student at ABS and as an artificial intelligence (AI) specialist with the city’s Chief Technology Office. Sukel’s supervisors are Stevan Rudinac, an Associate Professor at ABS, and Marcel Worring, director of the Informatics Institute and professor at ABS. ‘AI, the automation of decision-making processes that are currently still being run by humans, has a lot of potentials. Certain tasks can be automated and be done better using AI. AI is able to detect fraud in and recognise patterns, and it a fairer and better-substantiated way of working. It’s more efficient and less random – taking into account privacy, legislation and regulations.’
It was and remains a win-win situation. ‘I use the university’s infrastructure and network to build models. One needs massive computing power to develop a model for recognizing garbage bags. This is one of the projects I’m working on. Thanks to this cooperation, the city obtains new knowledge and the infrastructure to create such models, while the UvA gets interesting cases.’

Double focus

For his PhD research, Sukel works with service requests that contain not only text but also information in a variety of forms, such as image, location and time. By incorporating such modalities into the classifier, the model becomes more accurate. For example, a request regarding waste pick-up with an accompanying photograph could help determine if the request is for household waste, litter or bulky waste. This information is important to decide what kind of equipment is needed to clear the waste. A picture of a wrecked car, bicycle or graffiti with the text ‘can this be removed?’ would help clarify what kind of action must be taken.

For now, the city will use text input only. Sukel: ‘The authorities are under a magnifying glass. One has to be able to clearly explain how decisions are being made, and the city itself wants to be transparent. By incorporating all those other elements into models, it becomes more difficult to explain how they function.’ Besides, he says, city staff still need time to get used to the text classifier. ‘Many different departments and processes are involved. It would simply not be feasible for the organisation if the more complicated model were rolled out already.’

The PhD student considers doing research at both the UvA and the city a good match. ‘At the city, I’m trying to solve day-to-day problems happening right now. My thesis is geared more towards the longer term. At the UvA, I’m also investigating how one can keep this kind of problem from happening in the future.’

‘I want to improve the liveability of cities - Amsterdam and other cities as well.' 

Proactive and open

Sukel is an advocate for being proactive. Using models, he has developed, cities could look for possible improvements in the future. Thus, scanning cars that are currently used for parking enforcement, could be programmed in the future to also detect stray garbage bags and report these directly to administrators or to a suitable garbage truck passing nearby.

Sukel freely shares the models he builds for his research on developer forums. ‘I want to improve the liveability of cities - Amsterdam and other cities as well. People in different cities around the world are working on the same things, and this is a waste of resources… And if others would share their models, implementations and insights in turn, one can hope to get something back in the end.’

Youthful energy

Sukel, who officially works two days a week at the UvA and three at the city offices, plans to complete his theses in another two and a half years. Since his arrival, Amsterdam has welcomed numerous AI & Data Science students. Last year, 25 graduate students of the UvA carried out their research at the city offices, and so far, this year there have been 90 applicants. Sukel helps with the search for research projects and supervises several students. ‘Those masters students suddenly discover that they are already capable of quite a lot. For the city, it is also advantageous: there is a huge inflow of youthful energy. When these people start work, they might not have considered a job in the city; but some may stay once they have finished their studies because they find it exciting.’

He is in frequent contact with researchers abroad and has plenty of work for his PhD and his projects in the city in the years ahead. Once he has completed his thesis, he expects to spread his wings, he says. ‘Countries and cities continue to operate very much on a local level. But the technology I am working with is highly scalable and can very well tackle things on an international level.’

More information? Mail to