What is driving risk and why most apps DO NOT measure it although they seem to be interesting
Driving risk assessment is now of interest of :
. telematics solutions providers
. fleet managers (for prevention)
. car insurers (for prevention, and UBI)
A lot of applications then explain how they estimate driving risk.
But, by the way, what is driving risk ?
A lot of people seem to think that driving risk and accident are quite the same. If I observe accident, then I will learn and do stats and find a way to estimate driving risk. Sounds good ?
Unfortunately, it is completely wrong.
Let's fancy that : I drive in a city and I chose not to stop at red light (traffic lights) and not to stop at stop signs.
I do this 10 times and I don't have an accident (you all understand that it is possible). Then stats and machine learning will deduct that it is not dangerous to cross stop sings and red lights ... Although we all know that it is very risky to do that.
What happens then : you take risk, you watch carefully and you have luck, then you do not have an accident. But sooner or later, you will. Maybe in one month or two, but you will.
And you see, when the accident happens, you had more crossed stop sings WITHOUT accident than WITH accident. So you can understand that at the individual level, accident is NOT the measure of the risk you take (not at all).
When you cross a stop sign YOU DO TAKE RISK, whatever you have an accident or not !
So now you start to understand what DRIVING RISK means, and how it is not possible to get it by statistics or machine learning at the individual level (you have more counter examples than examples). So why do I say that CROSSING stop sign without stopping is dangerous ? Because I KNOW THAT.
This is called a KNOWLEDGE, and concerning the driving risk, there are experts all around the world that have KNOWLEDGE about driving risk. We'll come back to it a little further.
You also find Apps that "measure" the driving behaviour (example : lateral acceleration, severe braking) ... What the hell so you think that you can do with that ? estimate driving risk ? NO WAY.
Indeed, if you drive smooth and never brake ... and again cross a stop sign ... then you are very dangerous and you take huge risk (for you and for the others). It means that you need to know :
. driving behaviour
. context (where, etc ...)
It is obvious that the same driving behaviour on a disused airport, in front of a school, or approching a crossing road ... lead to different risks.
What does that mean ? Simply that all those Apps do not estimate driving risk (NOT AT ALL). Those that develop such Apps do not have a clue of what driving risk means ... But YOU, reader, now you get it !
I know it is anoying, but driving risk really means something, and you cannot estimate it if you do not have this information inside your data. Driving risk happens when driving behaviour IS NOT APPROPRIATE to driving context (infrastructure, traffic, visibility, ...). And if you do not know the context, NO WAY.
Another idea that we read from time to time is that it is possible to estimate driving risk by recording BLACK SPOTS LOCATIONS (on a map) ... sounds interesting again ? Unfortunately it doesn't work neither.
By the way, what is a black spot ? ... hum ... Black spot is location where a few (example : 3 or 5) driviers had an inappropriate driving behaviour those last years, they took risk and were unlucky then had severe accidents. But for YOU, with YOUR driving behaviour ... Does it mean something ? Will those black spots be black spots for YOU ? of course not : again, a location IS NOT dangerous by itself (as a driving behaviour IS NOT dangerous in itself). It is when driving behaviour IS NOT APPROPRIATE to context (for example : to instractructure complexity) that risk comes. And please do not argue that you've recorded the speed of cars that didn't have accident and speed of cars that had accidents on a large fleet ... Again, you may drive at 100 km/h on a tiny curve one thousand times without an accident (because you're a pilot) ... what will happen if next time a child is on the road in the middle of the curve ? ... ACCIDENT. Then again statistics CANNOT HELP ! and we know it is too bad but now that you read this paper, you understand.
Those big data bases of black spots (with speeds with and without accident) will lead to NOTHING although they seem to be a very good idea, don't they ? Because at the individual level, statistics DO NOT MEAN ANYTHING.
So the ONLY solution at the individual level is to measure how inadequate the driving behaviour is knowing the context (then "measuring" the context).
Fortunately, there is a module that does that perfectly : SafetyNex (smartphone App, or API that can be integrated inside another App or a device - Linux, Android, IOS). This solution, developed by the high-tech startup NEXYAD (based in France) is a KNOWLEDGE BASED ARTIFICIAL INTELLIGENCE (it took 15 years to extract knowledge from experts of 19 countries and validate it on 50 million km).
A solution exists, and it is easy to deploy on your Apps (smartphone) or devices (telematics, connected car, ...).
To read more : http://www.safetynex.net