Zircon Helps Portsmouth City Council Find Predictive Insights Hidden in its Traffic Data

Like many cities, Portsmouth is keen to reduce congestion and its many negative consequences. The average driver in the UK loses 61 hours of productive time per year due to congestion [1], equating to £7.5 billion loss to the economy, according to data provider INRIX. In 2022, 1 in 20 early deaths in Portsmouth were estimated to be due to air pollution, over half of which comes from road traffic [2].

Going back to around 2015, many of Portsmouth City Council’s roads have measured journey times between points, using sensors that track Bluetooth-enabled devices in vehicles. After amassing over four years' data, Paul Darlow, Portsmouth’s Traffic & Network Manager, wondered if it could be used to make rapid predictions about how traffic changes, which would help his team decide when to intervene to smooth traffic flows. He brought in Zircon to explore the possibilities.

The hypothesis was that a change in journey time at one point of the road could be predictive of traffic problems down the line. If so, it would help Paul and his team decide when to pull on their hi viz jackets and run out to direct traffic, and when to sit back and let things sort themselves out.

“The key point here” he says “is that we wanted to know what was possible, what wasn’t, and what other data we might need to make this better. At this stage we didn’t know what we had, so we didn’t want to blow a load of money on a predictive engine that might not tell us anything new. We needed a partner who understood both data and road networks, who could look at things intelligently and tell us what the opportunities might be, or tell us honestly if there was nothing in there of value.”

Solution Development

After cleaning up the raw data, Zircon began an exploration of the data sets, looking at changes in traffic patterns at specific points, and what happened further along the road network as a result. This involved applying machine learning techniques to explore what signatures of traffic changes were predictive of real problems, and how accurate they were.

“Accuracy and relevance matters,” says Darlow. “There are costs of intervening and costs of not intervening, so we want to know what’s going to happen – good or bad – early. We have quite a good instinct for this, so any tool needs to do something more than our own intuition in predicting whether traffic will get better, stay the same, or get worse to the point intervention is needed”.

The initial exploration found that journey time data at 15 minute intervals could predict how traffic would change further along the road network with 60% accuracy. A second exploration using five minute intervals offered over 80% prediction accuracy. “That’s the level where you start to get valuable early insights on whether or not to act, that humans might miss,” says Darlow.

John Jolly from Zircon adds “The skill of delivering projects like this is not just knowing how to use the technology, but understanding the problem. Machine learning is very powerful for finding hidden patterns in data, but only if you design it to look for the right patterns. The data points are not just numbers in a spreadsheet, they represent messy human behaviour on road networks with physical limitations. Lots of correlations are red herrings, so we really needed to draw on our experience of transport data to understand what’s normal and what isn’t, so we could point the machine learning in the right direction.”

“We needed a partner who understood both data and road networks, who could look at things intelligently and tell us what the opportunities might be, or tell us honestly if there was nothing in there of value.”

Paul Darlow

Traffic & Network Manager, Portsmouth City Council

The Outcome

The final project report showed that useful predictions could be derived from the established data streams. It showed how reducing time intervals improved predictive power, allowing Portsmouth to balance accuracy vs data processing costs in any future model design. And it made suggestions around how the model could be improved with vehicle count data, helping inform Portsmouth’s future data collection strategy.

Darlow concludes “The initial project did its job of telling us what we could do with the data, and gave us lots of other things to think about, all in a short period of time”

“We have other related concepts still to explore, but we hope that one day this work will lead to models that are integrated into traffic control systems which provide real-time advice to our experts. Beyond that, this could even form the foundation of self-regulating traffic signals that adjust to optimise traffic flow and respond to incidents. But the first step on that journey is to prove value in isolation, and that is what this project did.”

Why Zircon

“Zircon was the perfect partner for this work,” says Darlow. “They brought deep data analysis capabilities to the table, as well as understanding the world of traffic management, which led us to practical insights. The team size hit the sweet spot of giving us access to a variety of skills, whilst ensuring we were all close enough to work collaboratively and challenge each other to think in different ways. We were very happy with engagement, the quality of work, and the results.”

Zircon’s John Jolly concludes, “This project shows how machine learning can be used to take traffic data and use it to generate valuable predictions. But it’s just scratching the surface, and much of the value was in showing what else could be possible by adding other data sources. In fact since the project ended, we have seen even more possibilities for these sorts of predictions by bringing in MIDAS data on traffic flows in the UK strategic road network, which we think could spot problems on major roads before they even reach the city. As tech advances, and as you add in more data streams, and link up more models, incredible things become possible.”