Organisation: AISIN RoadTrace
Date of Publication: September 2024
Uploded to Knowledge Centre: 24 October 2024
This study suggests that connected vehicle data is more effective at predicting road collisions than traditional analysis of historic datasets.
The study, carried out by Aisin RoadTrace, compared the predictive efficiency of two methods:
- The traditional approach, which considers the locations of historic collisions as the most dangerous hotspots of the road network.
- An approach which considers the clusters of harsh braking events as future collision location candidates.
The aim was to determine whether future crashes can be predicted more accurately.
To conduct the analysis, the researchers used a set of data from connected vehicles driven over a large-scale road network in southeast England.
This included two million harsh braking events – defined as the threshold at which all the objects in a car that are not well stowed would fly off the seat.
In addition to this, they used a dataset of six years of validated collision reports.
The results indicate that harsh braking events bring comparatively more information than past collisions.
Clusters of harsh braking bring an improvement of 22% in prediction rate compared to an equivalent number of historical collision locations.
Additionally, researchers say they have shown that using harsh braking clusters allows for much quicker detection of future collisions – as well as having the ability to detect collisions that could not be predicted even using five years of historical road collision data.