Prediction Models for Auto Insurance Data using A Custom Loss Function
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
In the first half of 2023, major domestic property and casualty insurance companies in South Korea recorded automobile insurance loss ratios in the 70% range, leading to increasing calls for additional reductions in automobile insurance premiums in the latter half of the year. Furthermore, with digital insurance companies launching products with various riders, competition in the domestic automobile insurance market is expected to intensify. Therefore, the insurance industry recognizes the urgent need for precise automobile insurance prediction models that minimize the burden of automobile insurance premiums on the public while maintaining current loss ratios. In line with this trend, a specialized loss function customized to automobile insurance was developed. When calculating premiums based on predicted loss amounts, it is crucial to create a loss function that optimizes values under the condition that the sum of predicted loss amounts is greater than the sum of actual automobile losses. By applying this loss function to four machine learning models, we can see that using a custom loss function to automobile insurance significantly reduces the total premiums compared to conventional methods, thereby enhancing the competitiveness of insurance products.
Abstract
In the first half of 2023, major domestic property and casualty insurance companies in South Korea recorded automobile insurance loss ratios in the 70% range, leading to increasing calls for additional reductions in automobile insurance premiums in the latter half of the year. Furthermore, with digital insurance companies launching products with various riders, competition in the domestic automobile insurance market is expected to intensify. Therefore, the insurance industry recognizes the urgent need for precise automobile insurance prediction models that minimize the burden of automobile insurance premiums on the public while maintaining current loss ratios. In line with this trend, a specialized loss function customized to automobile insurance was developed. When calculating premiums based on predicted loss amounts, it is crucial to create a loss function that optimizes values under the condition that the sum of predicted loss amounts is greater than the sum of actual automobile losses. By applying this loss function to four machine learning models, we can see that using a custom loss function to automobile insurance significantly reduces the total premiums compared to conventional methods, thereby enhancing the competitiveness of insurance products.