3 min reading time

Car Crash Hotspots: Leveraging Machine Learning to Detect Probability, for Safer Roads


  • What is Machine Learning?
  • What is the service ‘Crash Probability Hotspots’?
  • What does Machine Learning bring to Proactive Road Safety?

Stay in touch!

Through our newsletter.

To find out how Michelin manages and processes your personal data, refer to our privacy policy.

How can we reduce the estimated 9,560 lives lost on the road during the first quarter of 2022 (7% more than last year according to the NHTSA)? Technology can provide a clearer and overall picture of driving behavior and driving events on roads. Machine learning goes even further. Drawing on vast amounts of data from a wide variety of sources, it can calculate the potential probability of a car crash occurring. Road safety experts substantiate their knowledge with data, calculations and facts that allow them to focus on and prioritize where and when to act, before crashes happen.

What is Machine Learning?

As DOT officials, road experts and local enforcement collaborate to address the problem of serious injury and deaths on American roads, technology and data are becoming an integral part of road planning and strategies that can help save lives. Professionals are turning to machine learning to understand and identify in much greater detail where problems are located on their roads.

Meeting an Accuracy Threshold to Deliver Key Insights

A branch of artificial intelligence, machine learning has become a supertool in a number of domains. Able to process vast amounts of data in far less time compared to humans, machines analyze in detail and can make estimations, predictions or decisions. Machine learning is where a computer learns from the data it’s given, using algorithms. Applying statistical methods, data scientists “train” the algorithms to find patterns, make classifications and reveal key insights. The calculations that ensue and the model that is created are compared to actual data, revealing how accurate the model is. The model is improved and perfected the more it’s practiced and with access to more data – similar to how a human learns.

Data scientists continue to optimize the model until it meets a desired threshold of accuracy. Patterns, classifications and probabilities become more pointed, specific and precise. Known for its use in dynamic pricing, weather forecasting and drug discovery – to name a few – machine learning is now also being applied to road safety.

Applying Machine Learning to Road Safety

So many things can affect safety on the road, from road design, to signage, visibility, weather conditions and road conditions. Drivers react to their environment as they experience it by making changes to their driving behavior to compensate. Some changes may be small. At other times, it may be that a car crash is extremely difficult to avoid given the context. Machine learning can analyze all the data and different elements and provide road experts with specific road safety facts.

What is the service ‘Crash Probability Hotspots’?

MICHELIN DDi has developed its own proprietary machine learning algorithms and model to indicate car crash probability. Drawing on past crash, traffic, driving and contextualization data, the model calculates possible outcomes and determines the probability of a car crash occurring. Helping road safety experts understand with much greater accuracy where risky zones are on the road.

Past crash data

Records of past crashes provide details of the types of collision, date, time and location, the road users involved and any other contributing circumstances. An important, if not always exhaustive source of information.

Traffic data

Average speed, number of drivers on a road, congestion, driving patterns throughout the day all give data about how a road is used by all drivers.

Driving data

A sudden or atypical change in driver behavior shows when drivers are reacting to their environment. Aggregating this data can bring to light repeated instances of harsh braking or acceleration, phone handling, speeding or suspected collisions that always occur in the same locations.

Other data sources

Weather conditions, road conditions, visibility, road works… Various factors can also impact driver behavior.

What does Machine Learning bring to Proactive Road Safety?

Crunching all of this data together can provide a much fuller view to road experts, along with valuable insights. Through machine learning and Michelin DDi’s proprietary algorithms, officials can detect, locate and assess atypical driving behaviors indicative of potential near miss incidents. From the starting point of car crash probability, experts can dig down to understand the severity of road safety hotspots, before exploring and analyzing the aggregated and contextualized behavior that is provoking them. All data can be consulted directly in any geographic information system.

MICHELIN  DDi’s suite of services includes 3 levels :

Crash Probability Hotspots

Locate, assess and rank, by crash probability indicator, hotspots in your transportation system. Modeling crash probability reveals all the locations on your transportation network where road safety is most compromised. Your expertise is substantiated by new technology, assessment and calculations that incorporate all facts, context, driving events and historical information about your road network. You prioritize your time and resources on those road locations that present the highest safety risk.

Near Miss Hotspots: Severity Ranking

Thanks to a statistical approach, this service allows road safety official to detect, assess and rank near miss hotspots according to their potential severity. You can then assess what could be causing the driver behavior, and prioritize hotspots according to your road safety plan and budget.

Driving Events

Zoom out one step further to understand what is causing the near miss hotspots in your network. Road safety experts detect, locate and count using GPS points where and when atypical driving events happen on their transportation network. Events measured include harsh acceleration, harsh braking, phone handling, suspected collisions and speeding. Based on millions of US connected drivers and vehicles, statistical analysis and algorithms identify which qualify as atypical events compared to threshold measurements. Events are aggregated and contextualized according to road type, weather conditions and time of day.

Safety in Numbers Straight from the Road

With MICHELIN Driving Data to intelligence solutions, you complement the official, reported data in crash reports with the actual and current driver behavior that is on your roads. Get a deeper and more accurate understanding of safety throughout your transportation network. You can then undertake a full and complete analysis of how your roads can be made safer for all road users, using driving events as KPIs to measure the before and after. Base your decisions on numbers that come straight from the road.

Leverage Mobility Data with Michelin experts ! Contact us