• 9 MIN

Using Real-Time Data Analytics to Make Public Transportation Smarter

Using Real-Time Data Analytics to Make Public Transportation Smarter

By SMU City Perspectives team

Understanding how real-time data can help with predicting passenger behaviour and reduce commuter wait time on the last mile journey

It might be an ambitious aspiration, but by 2030, Singapore hopes that 75 per cent of all peak-hour commutes will be conducted through public transport.

This vision of future urban mobility centres on fostering a ‘car lite’ society, where a smart, multi-modal public transportation infrastructure offers commuters an affordable and convenient alternative to private car usage.

To this end, the Singapore Government is strategising to promote extensive use of its excellent public transportation system, which may include driver-less or unmanned robo-taxis (also known as Mobility-on-Demand) that can ferry commuters to and from their doorstep to the nearest public transport node, be it bus or MRT. Furthermore, it aims to increase the adoption of “active mobility” through the construction of 700km of bike lanes, which includes cycling superhighways; progressively increase unmanned and electric vehicles, which leads to greater ability to automatically adapt to supply-demand imbalances over the course of a day; and improve last-mile commutes, which involves the building of sheltered walkways to ease one’s daily commute to and from the nearest bus-stop.

Currently, the Land Transport Authority (LTA) has been using sensors installed in public buses to gather information on buses’ real-time location and arrival times at various stops to improve their transport planning. It has also been collecting data from the fare cards of commuters to track commuter hotspots, and better manage bus fleets and commuter demand.

By optimising existing services through the use of data, smart cities can increase ridership on public transportation, thereby reducing the region’s carbon footprint and decongesting its roadways. Professor Archan Misra, Vice Provost (Research) and Professor of Computer Science at the Singapore Management University, talks about the need to predict passenger travel behaviour to improve the efficacy of the future ‘last mile’ and on-demand transport solutions (for both buses and trains) during his presentation, “Using Real-Time Data Analytics To Make Public Transportation Smarter” at the World Expo 2020.

The event was held in Dubai from October 2021 to March 2022, with an objective of catalysing an exchange of perspectives and inspire action to deliver real-life solutions to real-world challenges.

 Understanding the Commuter’s Journey

Understanding the Commuter’s Journey

Studies have shown that the overhead of the last-mile commute – the journey from and to a commuter’s residence to and from the nearest bus stop, plays a big role in a commuter’s reluctance to make the switch from driving. Even with the availability of sheltered walkways to make the last- or first-mile commute in comfort, rain or shine, commuters still lean towards private transport for a truly door-to-door journey.

Prof Misra believes that real-time analytics can reduce the friction in the last mile of an end-to-end public transportation journey which, in turn, will lead to a rise in its usage. Besides analytics, he also shares that AI applied to public transit transactional data can improve public transport resilience and convenience by reducing travel and wait times, and minimise the impact of mass transit disruptions.  More importantly, the data needed to support such real-time insights and decision making can be easily obtained from commuters’ smart card daily transactions that capture their start and final destinations of every trip for both trains and buses as they tap in and out.

Thanks to Singapore’s distance-based fare structure, which makes for a high adoption rate, “when we look at the electronic data trace by ‘tap-in, tap-out’ records, you’ll see that it is very representative and captures close to 100 per cent of travel records,” says Prof Misra.

To illustrate his point, he elaborated on two projects that his team and his departmental colleagues have worked on:

#1 – Predictive Disembarkation & Mobility-on-Demand (MoD)

This project focuses on the optimised use of unmanned vehicles that can reduce the wait times of commuters’ last-mile journey through the prediction of the “disembarkation count”—i.e., the number of people who will get off at a particular bus stop.

“While we have great tools to tell us when the bus will get there, we need to know how many people will get off,” says Prof Misra.

Predictive Disembarkation & Mobility-on-Demand (MoD)

To achieve this, Prof Misra and his team developed a mobility behaviour platform, called BuScope, to predict bus disembarkation, which fuses two key factors: Regularity, which is based on personal choice and pattern recognition (“If you take this route once, you will likely repeat the route,” he explains); and Conformity, which is based on the travel pattern of the majority of passengers on the route. In doing so, the results show that “the team was able to predict up to a 90 per cent success rate in determining the precise bus stop where a commuter will disembark, leading a low margin of error”, he says.

Thus, instead of a reactive approach, whereby a commuter would most likely need to wait for a while, while an unmanned vehicle would be summoned to the bus stop to continue on the last mile journey back home, the predictive and proactive approach would allow for the anticipatory repositioning of the MoD fleet so that an MoD vehicle will be waiting for the commuter once they disembark at the bus stop. This approach helps the last-mile fleet “to meet the evolving imbalances between supply and demand,” explains Prof Misra.

Prof Misra also shared that in simulation studies, driven by real-world commuting data,in Toa Payoh West, the predictive MoD results showed that the wait times were significantly reduced, by up to 75 per cent – or less than a minute. Such reduction also translated to a 10 per cent improvement in resource utilisation, allowing the commuting population to be served efficiently with a lower infrastructural cost.  

This, he adds, “is the power of real-time analytics”.

#2 – Prediction of Specific Travel Routes of Train Commuters

The second project, says Prof Misra delves into specific paths taken by passengers during their daily train commute.

“My colleagues monitored multiple routes from a source to a destination, and empirical evidence shows that people won’t always take the shorter route,” he shares. And this could be due to a variety of reasons, such as taking a longer route because it might be less crowded or involves a fewer number of transits. “There are different preferences on how people choose different routes,” he adds.

Prof Misra showcased the work by fellow SMU Professor of Computer Science Zheng Baihua and her team on analyzing smart card transactions generated by automated card fare systems during commuter journeys, and reported in her paper, TripDecoder: Study Travel Time Attributes and Route Preferences of Metro Systems from Smart Card Data.

Prediction of Specific Travel Routes of Train Commuters

To work around the obstacle of commuters’ varying route preferences, Prof Zheng and her research collaborators took a different approach and decomposed the extraction of route choices into two cascaded interference tasks – one, to infer the travel time of each link between two consecutive stations in a route that contributes to the total duration of any trip within the train network; and the other, to infer the route preferences based on historical trip records and the travel time of each travel link inferred in the earlier task. Their solution, named TripDecoder, is the first model that identifies and fully utilises trips within a metro system where multiple alternative routes are feasible, to better determine exact routes (including transfers) taken by commuters, even though such route choice behaviour is not directly captured by an automated fare collection system.

According to Prof Misra, the TripDecoder model allowed the team to estimate the travel time between each route segment. Each possible route is individually estimated, followed by an estimation of the route that passengers will take. The result, he says, is that “trip prediction errors or route prediction errors are reduced significantly”.  And thus, using TripDecoder, Prof Zheng and her team were able to both improve the identification of travel routes taken by each commuter and provide accurate estimates of the total journey time that a commuter would experience on different routes.  

What is the Data Used For?

So what are the applications for this kind of analytics, one might wonder? Prof Misra explains that using such inferences derived from real-time data is vital to making Singapore’s public transport more efficient and at the same time, entice commuters to use public transportation.

“You can get much greater prediction because you know where people will disembark,” he shares, as seen in his BuScope work. And with electronic data from transportation smart cards, you can even give people predictions on how crowded the train might be at a certain time, and even how crowded it will be six stations later, as seen in another of Prof Zheng’s work on passenger load, in her paper, Fine-Grained Prediction of Passenger Flow Inside Metro Network Via Smart Card Data.

More importantly, this knowledge can also help reduce the inconvenience experienced by commuters during breakdowns of such arterial services. More specifically, Prof Zheng’s work, which provides real-time insights on the composition of commuters in a train experiencing disruptions, allows for a more targeted, rapid deployment of “bridging bus services that best match where the final destinations of the commuters will be”, Prof Misra reiterates.

Towards A Car-Lite Society

In conclusion, the use of real-time analytics in a commuter’s public transportation journey can result in a high level of predictability, even on routes where a passenger has only made one past journey. As one notable example, the combination of individualised and flow-based predictions was able to predict a commuter’s disembarkation bus-stop with an accuracy of over 85 per cent, as well as a mean error of less than one to two bus-stops. This, in turn, can be used in a last-mile MoD system, where unmanned vehicles are pre-positioned to respond and anticipate commuters’ disembarkation demand more accurately, as seen in a 75 per cent decrease in commuter wait times.

Ultimately, Prof Misra believes that this use of real-time data analytics will allow public agencies to view mobility data as not just a policy planning resource, but also as an enabler of a new class of “proactive” or “anticipatory” smart city services in Singapore’s vision of a car-lite society.