scroll

Safer Roads with Machine Learning

The challenge

Road accidents are a global problem – fatalities occur daily in almost every country. Yet, in some way or another, we’re all reliant upon these roads to go about our lives. We collect masses of data about these accidents and use this information to upgrade roads where crashes often occur. What if instead we could use predictive algorithms to prevent these crashes before they even happened?

In this article

  • The impact that road crashes have on our communities.
  • What governments are currently doing to make our roads safer.
  • How we can use machine learning to help these governments make more proactive investments in road safety.

Transport is an integral part of everyone’s life. Whether you drive a car, take public transit, ride a bicycle or catch an Uber, every day you spend a portion of your time on the road. Statistically, that may be the riskiest thing you do all day. In the year 2013, the World Health Organisation estimated that there were 1.25 million deaths because of road crashes—the equivalent to one death every 25 seconds.

 

“It is easy to forget your personal attachment to a statistic or a number but consider this: how many people do you know that have been involved in a road crash?” says Ryan Wong, a transport planner in our Singapore office. “It’s our goal as traffic city planners to make that number zero.”

Road fatalities and injuries are a problem every country faces. Crashes have a terrible effect on families as well as the economy. It has been a problem we have been grappling with since the advent of mass motor vehicle adoption. Most government agencies recognise this and are constantly investing in making their roads safer for all uses. The problem is that most of these investments rely on historic crash data. Something needs to go wrong before an intersection or mid-block is flagged for improvement.

Ryan believes this doesn’t have to be the case. “What if instead we could make our roads safer by fixing problems before they even become problems?” he says.

To do this, he proposes using modern machine learning to develop a tool capable of identifying future problem areas. Most roads aren’t designed unsafe from the get go, they simply become that way as the road’s use changes over time. Since a wealth of data exist around each accident, Ryan believes we can use this to teach machine algorithms to identify trends. These algorithms can then help us determine where infrastructure improvements might be needed soon, as traffic volumes begin to increase along a previously uncongested corridor. By investing in upgrades early on, we can save time, money and—most importantly—lives.

Road accidents have been a problem we've been grappling with since the advent of mass motor vehicle adoption. They have a significant impact on families as well as the economy. Fortunately, Ryan thinks we now have the data - and the technology - to get better at preventing them before they occur.

Consider two different road upgrade strategies. The first involves investing $10 million to upgrade an intersection where 50 crashes have occurred over the past five years, such as those identified by the Australian Government’s Blackspot Program. The second involves spending the same $10 million on seemingly minor upgrades like line-markings or road treatments at 20 different intersections, each which has seen five crashes over the past year but is beginning to see an increased traffic demand. In a perfect world, all 21 intersections could be upgraded. But budget constraints dictate that under current processes the need to improve the ‘Blackspot’ intersection would outweigh the others. Ryan believes his predictive algorithms could help governments weigh those two strategies more objectively. Our surging population—and our increasing dependence upon motor vehicles as our suburbs sprawl—mean it is no longer sufficient to rely entirely on the past to predict the future.

Ryan would like to develop his process into a digital tool that can pull in third-party crash data, assesses it with respect to variables like speed, traffic rates, and road geometries, and deliver insights on a graphic user-interface. These insights can then be used by decision makers to better understand why crashes have occurred in a certain area, where future crashes may occur, and how they might proactively intervene.

Ultimately, Ryan hopes his tool can help government agencies make more targeted investments, resulting in safer, more reliable roads.

“The output of our study is a tool but the real goal for this study is to save lives,” Ryan says.

Roads are rarely built to include unsafe turns or intersections. Instead, these areas become unsafe as traffic rates and road usages change overtime. What if we were to use predictive algorithms to detect and prevent this? By making informed investments in road upgrades early on, we can save time, money and—most importantly—lives.

Findings

  • Good data are hard to come by – crash statistics are recorded by an officer on-site. Therefore the characteristics of the crash can be subjective and require double checking.
  • So many factors in crash statistics are highly variable. Crashes can occur for reasons often not captured in the data.
  • Despite the imperfect data, machine learning algorithms can identify patterns we wouldn't have otherwise seen through traditional analysis.

This story was written by Jeff McAllister and Ryan Wong, as part of the Research Review. This series is produced by the Arup Australasia Research team; Alex Sinickas, Bree Trevena and Jeff McAllister with contributions from Sheda and Noel Smyth.

Lead Arup Researcher

Ryan Wong
Ryan is a transport engineer in our Singapore office.

Ask Ryan about:

  • Getting involved in the project - through contributing data.
  • Agency involvement – we are working with both Transport for New South Wales and New South Wales Roads and Maritime Services, but are looking for more partners and data.
  • How autonomous vehicles fit into this project.

LEAD Partner RESEARCHER

Research TEAM

No items found.

Have a problem or a project?

We work with industry partners, governments, universities, startups and community organisations. We do this through research partnerships, and as consultants and facilitators for foresight, research, storytelling and technical writing workshops.

research with us