How Scoring Works in the Car Sharing Industry. Part 2. How to Identify an Aggressive Driver in 5 Seconds.
In the previous article, we considered a scoring algorithm based on rush accelerations and sudden decelerations. The scoring results that this algorithm produces have a weak correlation with accident probability. In this article, we will take a look at more advanced driving style analysis algorithms that are based on speed, engine RPMs, and accelerometer indications.
We will analyze the data of 250,000 routes and 43,000 car-sharing users. 250 of them were responsible for accidents. Remoto telematics devices were installed in the vehicles under analysis. In the article, we will consider two groups of drivers — aggressive and safe drivers. Aggressive drivers are defined as those who repeatedly violate traffic rules.
Well, first, let’s analyze the average speed and average engine RPM of users who did not get into accidents and the hit-and-run drivers.
Blue and orange are used at graphic 1 and graphic 2 to specify the indicators for users who did not get into an accident and the hit-and-run drivers, the dotted lines mark the medians. You can see that the average indicators for users differ slightly. This indicates that both types of drivers drive most of the route alike.
After we selected 10 drivers: 5 who drive aggressively and 5 safe drivers. Analysis of the data from the routes driven by these user groups revealed one interesting feature. The aggressive drivers accelerated quickly and attained higher engine RPMs during the first 5 seconds of the trip. In addition, aggressive drivers brake from greater acceleration in case of car stop (at traffic lights or at the end of a trip). We analyzed the following values for speed, engine RPM, and acceleration:
· Maximum value
· Maximum change in value
· Average standard deviations
· Average value
· Median value
It is clear from our test sample that the maximum RPM for aggressive drivers was much higher at the beginning of the route than the RPM of other drivers. The same was true of peak acceleration. As a result, we decided to calculate these values for all 250,000 routes, splitting the data into two groups: the hit-and-run drivers and ordinary drivers.
Figure 3 shows the distribution of the maximum engine RPM. Orange is used to indicate trips by users who caused accidents, and blue indicates trips by all users. The dotted lines indicate the median values. You can see a clear rightward shift in the line for the drivers who are responsible for perpetrating accidents, i.e., during the first 5 seconds of the trip, these drivers rev their engines to a higher RPM than others. A total of 36% of drivers from this group achieve engine RPMs are higher than 4,800 at the start of their trip, at the same time only 9% of all drivers in the sample ever reached that RPM.
A similar situation can be seen with peak acceleration. Figure 4 shows the distribution of peak acceleration values. The peak acceleration of the hit-and-run drivers is higher than it is for all other drivers.
Now we will consider not only the start of the trip but all five-second intervals beginning after a full stop and five-second intervals preceding a full stop. For every 5-second interval after the car accelerates from a stop, we calculate the maximum value of the engine RPMs and peak acceleration. Then we find the maximum values for each driver.
The maximum RPM indicator demonstrated the largest deviation at the start of the acceleration. The peak acceleration indicator demonstrated the largest deviation for stops.
As a result, it turns out that 80% of the hit-and-run drivers have engine RPM values in excess of 5,000, whereas only 20% of all drivers do. So, in other words, 80% of the drivers who are responsible for causing accidents to include 20% of all drivers. In addition, 52% of drivers who have caused accidents had engine RPM values in excess of 5,800, whereas only 10% of all drivers in the sample have achieved such values. The same thing can be observed for peak acceleration in case of stop. However, the difference is less visually clear. A total of 60% of drivers who got into accidents showed acceleration values of more than 120, at the same time only 20% of all drivers showed that rate of peak acceleration when braking to a stop.
We are considering the maximum values because of the strong variation of the indicators for aggressive drivers. The deviation of the data for one aggressive driver is always higher than that for a safe driver. In other words, the safe drivers usually drive the same, whereas aggressive drivers may sometimes drive safely but at other times extremely aggressively. It’s hard to say why this behavior exists and what affects it (reasons could include driver mood or rush, etc.).
We considered the starts and stops of each rout and obtained a clear deviation for drivers who got into accidents relative to the entire group of drivers. Drivers with an aggressive driving style are most often responsible for road accidents. 80% of the perpetrators of traffic accidents can be classified as aggressive drivers in terms of maximum engine RPM achieved when the car starts to accelerate. Car sharing operators should pay attention to this group of drivers because accidents are always a financial loss for these companies.
About the turns as important indicators of the routes I will talk about in the next article.
