dc.description.abstract | Taiwan′s "Ratio of Assets of Insurance Industry to Total Assets of Financial Institutions" has risen from 17.36% in 2004 to 36.32% in 2020, which is the highest point in recent years. This shows the importance of Taiwan′s insurance industry. Insurance is not only a financial tool to facilitate socio-economic activities, but also plays a vital role in stabilizing social system. According to the seminar on the prevention of insurance crime held by the "Institute of Financial Law and Crime Prevention" in 2020, it is pointed out that the estimated loss of insurance crime in 2019 will reach NT$10.5 billion, and it is also pointed out that foreign experience accounts for about 10% of the claim amount. , this potential insurance fraud and adverse selection risk is an important subject of risk control for insurance companies.
This study uses the underwriting cases of Company A in 2018 as the data source, and takes the non-acute medical claims cases within two or three years after the policy takes effect as the target variable. Analyze and discuss six aspects of customer basic information, insurance information, financial information, physical condition information, agent information, and agent claim settlement rate, and use machine learning technology to try eight different classifiers to build underwriting risk prediction models and experiments. From the experimental results of the seven classifiers, it can be seen from the independent variables that the number of medical insurance, the type of primary insurance, the code of the primary insurance and the number of riders under the policy are the best predictors. The performance of the prediction model is based on the gradient boosting machine, which has relatively stable and high prediction ability, and its AUC reaches more than 0.71. For CA scores, logistic regression is the best, reaching above 0.612.
It is hoped that the results of this study can provide a reference for the case company to construct an underwriting risk assessment mechanism for risk classification based on data in the future, and provide a more efficient underwriting assessment and underwriting risk classification through the underwriting risk prediction model. So as to improve the underwriting risk classification and efficiency, speed up automated operations, and improve customer experience and satisfaction, thereby enhancing customer’s stickiness and loyalty with the company. | en_US |