Crowd prediction

Challenge

Zandvoort is one of the most popular beach towns in the Netherlands. Every summer many people from Amsterdam and far beyond visit the beach, often on a day trip. Although the beach itself can be crowded, the real bottlenecks at Zandvoort are the entry points to the beach and specifically the walking route between the train station and the beach. 

Zandvoort knows that problems with crowds – from nuisances like littering to unsafe situations even – grow quickly with crowd size. Of course the amount of visitors highly depends on the (predicted) weather, day of the week, school holidays, etc. Therefore, the municipality asked the collaboration of Lynxx, 9292, and HERE to predict several days in advance when to expect most people, focussing on Saturday and Sunday as the most crowded days of the week. Zandvoort also wanted help to steer visitors towards arrival times that are less crowded.

Our solution

We built 3 modules: prediction of the amount and arrival times of beach visitors; a real-time monitoring tool of the latest prediction; and a messaging system to alert and steer prospective travelers through the 9292 journey planner app.

Lynxx is a longstanding partner of 9292, which is the Netherlands’ most widely used independent provider of travel advice for public transport. We used the trip planner data of 9292 starting from 7 days in advance to gauge the amount of people planning a trip to the beach of Zandvoort. We also partnered with HERE WeGo to use Floating Car Data from Garmin navigation as well as the Flitsmeister-app in order to assess visitors arriving by car.

Next to this travel information we collected the weather prediction and other explaining variables in order to train an Artificial Intelligence algorithm. We used historic days with known amounts and patterns of visitors at Zandvoort as a training sample. With this, we could provide a clear picture of what the crowd pattern could be expected to look like on a given day. As the day of interest grew nearer, the predictions got better and better, as more people were planning their trip using the 9292 journey planner app, and the weather forecast got more accurate. Also during the day the prediction was continuously updated, based on actual visitor numbers, new planning requests, and road trips.

We also provided the municipality with a dashboard to monitor the situation leading up to and during the busy weekends. In this they had a real-time view of the current and expected visitor numbers. Flexible trigger values could be set to either alert personnel to manage the situation, and to alert travelers of the crowds to expect. 

To steer travelers away from the most crowded arrival times, we fed back the predicted pattern of visits to the 9292 journey planner. This meant that if someone planned a trip to Zandvoort beach, arriving at a busy time, the user would see a banner warning for crowds and suggesting to adjust travel plans accordingly. When the user clicked on the banner, (s)he could see the predicted crowding pattern and start planning the trip at a different time with less crowds expected.

View of apps to support travelers when planning a trip to Zandvoort. From left to right: the 9292 travel advice, including banner warning for crowds; the 9292 page with the expected crowds over the day; the HERE WeGo app showing crowds

Results

Our AI-model proved to predict the expected crowds well. Especially several days in advance this improved the view of Zandvoort on the amount of visitors to expect in the upcoming weekend. This meant that they could respond more proactively instead of waiting to see how the situation would unfold.

With our feedback mechanism, we reached 7000 people that used the 9292 journey planning app each weekend day. Of these, 22% of travelers clicked on the message to look for alternative times to arrive. Survey results of the users showed that 79% of participants appreciated the heads-up about when to expect crowds, and 29% indicated that they adjusted travel plans accordingly.