Road accident prediction company Predina uses data to predict the location of high-risk driving routes and alerts drivers so they can avoid them or take extra caution. Co-founder and CTO Meha Nelson said that the premise of Predina is that the external environment is what causes an accident, not just the human driver.
“If you are in an area which has a high risk, one mistake could turn into an accident but if you are in another area, you might have a near miss or you might get away with it,” she said.
The data start-up initially worked with one supplier of natural gas and assessed that company’s accident data from their fleets trucks, in addition to UK government traffic data from the past 10 years.
In December 2016, Nelson alongside co-founder and CEO Bola Adegbulu decided to combine her experience in computer science and AI, and his experience of having founded an automotive start-up to create a technology that would benefit the social good of society. “We wanted to create an impact and touch the lives of human beings and basically solve a problem,” she said.
In 2018, in Great Britain alone, there were 160,378 reported road traffic accidents and 1,782 reported road deaths. The aftermath of accidents lead to roads being closed, traffic being diverted and increased congestion – congestion in the UK is estimated to cost the economy £7.9 billion per year.
The industrial gas company that Predina partnered with could suffer a loss of £1 million from one major accident and if one of their drivers hit a bridge or a lamppost the company will “have to pay a lot of money to the government,” according to Nelson. Fortunately the data that the gas company provided to Nelson and her team was anonymised and so they did not have to do any additional processing. The enforcement of sanctions for breaching GDPR did however mean that data was not as easy to get hold of as it had been previously.
“Before they were able to send us data via email, it was simple. After a lot more people had to get involved because basically they had to write a script in which they anonymised a lot of things, a lot of data would be removed and then it would reach us,” said Nelson.
Once the data was received, the Predina team would have to put it into a machine readable format. The accident reports were hand-written with drivers often putting all the information in the description and leaving the remaining columns blank. Nelson had to bring in freelancers, students and interns to read the information and fill in the appropriate categories. “It was a semi-automated process,” she said. “We used a bunch of natural language processing and clustering algorithms as well but there was a human in the loop, largely because the data was very unclean.”
To create those algorithms and “drive the development of the data science side of things” Nelson worked with an extreme risk modelling expert from the University of Edinburgh and the Alan Turing Institute, a geospatial data expert from the National Physics Laboratory as well as a machine learning post doc with experience in the industry.
Upon A/B testing the algorithm, with 150 drivers being assisted by the solution and another 150 driving without it. Those who were assisted were given real time feeds in the form of audio alerts through their navigation device informing them when they were about to enter or leave an area that had a high risk of accidents. In addition the managers of the drivers were sent a PDF of the drivers scheduled to pass through high-risk areas informing them of the change in risk over the 12 hours from 6am to 6pm.
The PDF would be printed out and given to the driver.
“We benchmarked over previous year and we saw that the drivers we provided the solution to, there was a 25% reduction in accidents, which again could be a fluke but with the drivers who we did not provide the solution there was a 41% increase in the number of accidents compared to previous years,” said Nelson. In addition, Predina was able to predict with 70% accuracy.
There is a clear benefit to the drivers and the company in terms of time and money saved from not having collisions but this is also good for other road users. Furthermore, the information on accidents hotspots can be used by emergency response planners who can use it to determine the best place to station ambulances and where to put dynamic road hazard signs. “This was a use a use case both in the UK and in Australia. These are longer term implications but we are having conversations,” Nelson said. The upside for Predina was that it received investment from the gas company.
Although she left the company in May 2019, she learnt a lot from leading her first start-up which was part of the Data Pitch innovation programme accelerator, an initiative that facilitated start-ups getting access to private and public sector data. They include how to manage financial and human resources, so how to manage money and whom to hire, as well as how to set and achieve milestones so that she and her co-founder could keep each other accountable.