Predicting disruption fallout with passenger movement data


518 words / 2 minutes

passenger movement
public transport
mobile data
data visualisation

Last month I attended DfT Hacks, the first external hackthon by the UK's Department of Transport. Over 48 hours the participants tackled problems faced by the Government and the transport industry.

On Friday we met at the Department of Transport in Westminster. We were joined by a host of organisations including HS2, Transport for London, O2, Reading Buses and the Department of Transport. Each organisation presented the problems they face and provided datasets and APIs to help solve them.

Our team chose to focus on passenger movement1. A transport disruption can be a stressful time for both operator and passenger. As a disruption occurs, passengers look for an alternate means to continue their journey. This can impact the surrounding transport network but what if you could predict where disrupted passengers go? Could you add extra buses to a route or increase the number of hire bikes in the area? After a quick brainstorming session we had a rough plan: Visually map how passengers move during and after transport disruptions.

Mobile network data can be used to map the movement of human traffic and we were able to source a dataset like this for the city of Milan, Italy. The data was depersonalised - providing only the number of mobile devices connected to a network within a certain area during a certain timeframe. For example, in the last 5 minutes there were roughly 60 mobile devices within this 500 metre square. Azienda Trasporti Milanesi (ATM) is responsible for the majority of public transport in Milan. We used their twitter account and Microsoft's Cognitive Services to translate all the tweets during our timeframe and pinpoint a substantial disruption to visualise a shift in passenger movement.

On Saturday our developers began structuring the data into referenced grid squares. I used this time to explore a sequential colour scale2 that could visualise the number of people estimated to be within a 500 meter grid square. I decided on using one colour, increasing or decreasing the brightness, with dark intense colours representing higher values. With the data transformed into grid squares we overlayed it onto a map of Milan and watched, with tired eyes, as it danced around the map.

On Sunday we put our tool to the test as we presented it to the judges. We highlighted grid squares containing a metro station and queued up the data before, during and after a substancial disruption. As the disruption time approached we began to notice a shift. The human traffic in the disrupted grid square started to decrease and we were able to visualise and monitor it as it flowed through adjacent grid squares to reach an alternative station. We'd just visualised passenger movement during a disruption. Interestingly, we were able to see that the disruption and subsequent fallout began 20 minutes before the first public announcement.

Our team received the prize for 'Best Use of Mobile Data'.

  1. Passenger movement is the flow of human traffic through a transport network.

  2. A sequential colour scale is used to present data that is continually increasing or decreasing over a period of time.