Car insurance companies often rely on simple mathematical models, such as linear regression, to calculate actuarial risk. These models, first used in the 1970s, were designed when computational capabilities were limited and therefore had a simple calculation method. As a result, they were straightforward to use. Another commonly used method is Generalized Linear Models (GLMs), which were introduced in 1972. However, these models also have limitations and may not accurately reflect the unique risks faced by each driver. As a result, these methods often result in an inaccurate assessment of risk of drivers.
The cost of relying on outdated methods of calculating actuarial risk is a reduced ability for car insurers to accurately differentiate between good and bad drivers. This results in higher premiums for safe and responsible drivers, who effectively subsidize the risk posed by less safe drivers. In most cases, this will lead to paying more for insurance coverage that may inaccurately reflect your individual risk profile.
The latest and greatest method in actuarial risk calculation is the use of Neural Networks (NNs). NNs are highly flexible and accurate models that offer a more personalized approach to car insurance. NNs utilize large amounts of data and sophisticated algorithms to provide a more accurate assessment of risk. This innovative approach is a significant improvement over methods such as linear regression and GLMs and, in our opinion, represents the future of car insurance.
However, there is a major drawback to using NNs in actuarial risk calculation. Unlike methods, such as linear regression, NNs are considered “black boxes” as there is no way to understand how or why the model arrived at its conclusions. This lack of transparency can make it difficult for the insurer to understand why drivers are being classified as high or low risk, or for insurers to justify their premium pricing.
In addition, the lack of interpretability in NNs also presents a significant risk to insurance companies. If, for example, the NN were to flag all drivers from Morocco as high risk, this could result in a major public relations disaster for the insurance company and potentially even face regulatory action in the US and EU. The use of NNs in actuarial risk calculation is therefore a double-edged sword, offering great potential for more accurate assessments, but also presenting significant risks to the reputation and legality of the insurance company.
As a technology company, we recognized the importance of being able to explain the reasoning behind our models, even when using cutting-edge techniques like NNs. That’s why we’ve made it a priority to develop methods that allow us to maintain the accuracy of NNs while also providing transparency and interpretability.
To achieve this balance between accuracy and interpretability, we utilized a combination of advanced tools and innovative techniques. Drawing on the expertise of our team, which includes some of the brightest minds in the field, we were able to develop a solution that leverages the power of NNs while also providing the transparency and accountability that is essential in the insurance industry.
Instead of attempting to understand the specific reasoning behind the NN’s decision-making, we took a different approach. We started by defining well-separated groups and then trained the NN to work within these clearly defined boundaries. By inverting the process, we were able to ensure that the NN was making decisions based on a set of clearly defined criteria, rather than relying on opaque, potentially biased algorithms. But wait, you might say, did you just solved one problem by creating another?
To achieve this, we utilized t-SNE, an algorithm that is capable of separating data into distinct groups while maintaining the relationships between those groups in a way that is understandable. However, t-SNE has limitations and cannot be used for making predictions which as we have established, NNs can do very well.
This method brings together the best of both worlds. On one hand, we are able to break down the data into understandable groups, where the characteristics of each group are easily defined. On the other hand, we are able to accurately place each driver into their own group using NNs. With this approach, we are able to maintain the benefits of both t-SNE and NNs making NNs fully transparent!
In conclusion, our innovative solution to combining t-SNE with NNs in actuarial risk calculation overcomes the drawbacks of using NNs alone. By breaking down the data into understandable groups using t-SNE, our method provides transparency and interpretability, making it easier for the insurer to understand why drivers are being classified as high or low risk and to justify premium pricing. Furthermore, our method eliminates the risk of a public relations disaster, as the use of t-SNE ensures that no group of drivers will be flagged unjustifiably as high risk.
And for you, the reader, it is important to know that our innovative combination of t-SNE and NNs in actuarial risk calculation results in more accurate assessments and reduces the price of insurance.
Source: Kaso2go
