How Scoring Works in the Car Sharing Industry. Get a Quick Overview of Popular Tools
Car sharing is one of the most dynamic areas in the automotive business industry. Data collection goes hand in hand with the development of car sharing. And now, like banks, car sharing companies implement a scoring system. However, as you may have guessed, the scoring system isn’t based on your credit history, but on your driving history. In addition to ability to pay, driver’s license validation, and traffic tickets, the system is designed to predict the chance that a driver will get into an accident.
We should point out that, in car sharing, a car is generating revenue. Given this approach, it is important to use the car as efficiently as possible and to avoid any downtime. If a car is involved in an accident, then dealing with the insurance company, coming to an agreement with a mechanic, ordering parts, and getting the car fixed can take a time. Scoring can predict potential accidents and provide the feedback to customers about the risks of unsafe driving based on its.
In this article, we will address to the logic behind how scoring algorithms for car sharing users work with the focus on algorithms for age and driving style. We will present how scoring works based on anonymous data of 50,000 users and 260,000 routes. In addition, we have also used data about 220 accidents recorded in a one of European capital city.
As we mentioned above, car sharing providers need to protect their cars and made a lot of money out of this.. Car sharing platforms are therefore worked to collect all the data about their cars and about their rides. Car sharing vehicles gather telematics data during every trip. This data includes the car’s location at regular intervals of no more than one second and its indicators at these locations (speed, acceleration, door and window status, etc.).
Age
When the customer signing a contract with a car sharing provider, a driver must indicate his age and driving experience. We can do the following graphic based on this data.
Figure 1: User’s Age
Figure 1 shows a graphic of the car sharing users ages. The users’ ages are presented along the X-axis, and the number of users is presented along the Y-axis. The dotted line shows 30 years as the median value .
Now let’s present the breakdown of the the hit-and-run driver’s ages.
Figure 2:The hit-and-run driver’s ages
Figure 2 shows graphic a of the breakdown of the hit-and-run driver’s ages. The users’ ages are presented along the X-axis, and the number of users is presented along the Y-axis. As before, the dotted line represents the median age — 26. This makes it clear that among users who are younger than 26 are more the hit-and-run drivers than others.
The graphic shows that half of all accidents were caused by a group that makes up a quarter of all users (users aged 26 or younger). At the same time the group of users over 30 makes up half of all users but was responsible for only a quarter of all accidents. Also we can see that the number of accidents depends on age.
So we are looking at that 26 years old users or younger are four times more likely to get into an accident than users who are over 30. Thats mean that car sharing providers should highlight to the younger group of users.
Figure 3: Ratio of Number of Users to Number of Users with an Accident
In the chart in Figure 3, age is presented along the X-axis, and the ratio of the number of users with an accident to the number of users of a given age is presented along the Y-axis. The chart makes it clear that users who are 26 or younger get into accidents more often, and also we can see the same case for 47–49 years old drivers.
Driving Style
Things get more complicated when it comes to driving style. At the automarket there is a long-standing model for definition of driving style as parameter showing the sensitivity for accelerations and decelerations.
A is the number of rush accelerations per trip, and B is the number of sudden decelerations per trip. In the chart in Figure 4, we can see the the breakdown of the sum A+B on the sample routes.
Figure 4: Distribution of Sudden Accelerations and Decelerations
We should point out that a trip can last 15 minutes or five hours, so we need to focus on the time or distance of the trip for choosing the scoring parameters. D is a trip duration in minutes. Now let’s count the number of sudden accelerations and decelerations per trip minute, i.e. (A+B)/D. We end up with the asymmetrical breakdown shown in Figure 5 with the values on the left drop much more quickly than those on the right. Unfortunately, most statistical methods don’t work for distributions with drastic deviations. In these cases, you can usually apply a logarithmic transformation, which frequently transforms asymmetry into symmetry because it allows you to stretch out the scale around the area of zero.
Figure 5: Distribution of Accelerations and Decelerations per Trip Minute
If we find the logarithm of this function, we get Log ((A+B)/D). In conclusion, the distribution is very similar to a normal one: Figure 6.
Figure 6: Logarithm of the Number of Accelerations and Decelerations per Trip Minute
Scoring models for driving style are usually build on the basis of this function. Let’s try running all trips by every user through this function.
Figure 7: Each Driver’s Worst Score
Figure 7 shows the worst trips taken by each user. The blue bars are for all users, the orange bars are for users with an accident, the blue dotted line is the median for all users, and the orange dotted line is the median for all users with an accident. The median of users with an accident is clearly shifted to the right, i.e. users with an accident stand out from the rest of the group when the data is viewed in this way.
Let’s build a scoring model based on this method: the histogram in Figure 8. The scores of all users are marked in blue, and those of users with an accident are marked in orange. Scores range from 0 to 10, with 0 being the worst and 10 being the best. The median scores of the two groups of users are shown with dotted lines. The average score of users with an accident is about 4, and the median score of all users is 5. 80% of users with an accident have a score that is below average. In other words, 80% of users with an accident are worse-than-average drivers.
Figure 8: Scoring Results
A similar model is typically used when scoring drivers based on telematics data. The results can be used as a basis for restricting access to premium vehicles or to the service in general. However, it is not the only or optimal model for all cases.
